Systematic identification of pathological lamin A interactors

INTRODUCTION

The nuclear envelope (NE) defines the boundary between the nucleus and the cytoplasm (Macara, 2001; Dittmer and Misteli, 2011). The nucleoplasmic face of the NE is lined by the nuclear lamina, a dense protein interface connecting the inner nuclear membrane and chromatin (Stuurman et al., 1998; Dittmer and Misteli, 2011). Major structural constituents of the lamina are the intermediate filament proteins lamins A and C, encoded by the LMNA gene (Fisher et al., 1986). The A-type lamins are ubiquitously expressed in most adult somatic cells and assemble into higher-order filaments. Together with the B-type lamins, expressed from separate genes, lamin complexes contribute to nuclear structure and influence nuclear organization (Dittmer and Misteli, 2011).

Although the most-studied functional aspects of A-type lamins are the structural roles they play during interphase, lamins A/C are increasingly recognized as mediators, and possibly regulators, of diverse nuclear processes (Gonzalez et al., 2011). They are implicated in higher-order genome organization, since they physically interact with chromatin, and cells deficient for lamins A/C show a loss of peripheral heterochromatin, ectopic chromatin condensation, and mispositioning of centromeric heterochromatin (Taniura et al., 1995; Sullivan et al., 1999; Nikolova et al., 2004; Galiova et al., 2008; Bruston et al., 2010). Lamins A/C may also play a role in DNA repair as cells expressing mutant forms of lamin A exhibit a delay in recruitment of repair factors to sites of DNA damage and show increased levels of double-strand breaks (Liu et al., 2005; Manju et al., 2006). Lamins A/C have also been linked to signal transduction pathways involved in differentiation and homeostasis by regulating the activity and availability of proteins within the mitogen-activated protein kinase– and RB-cell cycle pathways (Gonzalez et al., 2011). They are also implicated in transcription and DNA replication, possibly by aiding assembly of the transcription and replication machineries (Kumaran et al., 2002; Spann et al., 2002; Zastrow et al., 2004).

Many of these activities rely on lamin A/C–protein interactions. Lamins A/C associate with B-type lamins and homopolymerize (Ye and Worman, 1995). In addition to lamin–lamin interactions, lamins bind to proteins anchored in the inner nuclear membrane, such as thymopoietin (TMPO; LAP2) or emerin (Zastrow et al., 2004). Together with SUN1 and nesprins, lamins form the LINC (linkers of the nucleoskeleton to the cytoskeleton) complex, which physically connects the nuclear membrane and lamina to the cytoskeleton (Crisp et al., 2006). Lamins A/C also interact with nucleoplasmic proteins such as barrier-to-autointegration factor (BAF), which prevents viral integration, and chromatin remodeling complexes, including the nucleosome remodeling and histone deacetylation component retinoblastoma-binding protein 4 (Montes de Oca et al., 2009; Pegoraro et al., 2009). In some cases, interaction with nucleoplasmic proteins serves to sequester proteins to the nuclear periphery, thus reducing the nucleoplasmic concentration. For example, lamin A physically interacts with the transcription factor Fos and retains it at the nuclear lamina, leading to suppression of activating protein-1–dependent transcription (Ivorra et al., 2006; Gonzalez et al., 2008).

Mutations in the LMNA gene cause a heterogeneous set of human diseases. More than 300 LMNA mutations have been identified and causally linked to >15 distinct diseases, which are referred to as laminopathies (Szeverenyi et al., 2008; Dittmer and Misteli, 2011). The laminopathic disease mutations are distributed throughout the LMNA gene and show little correlation to disease type (Szeverenyi et al., 2008; Dittmer and Misteli, 2011). A prominent hallmark of laminopathies is their high tissue specificity. Among the laminopathies, Emery–Dreifuss muscular dystrophy (EDMD), limb girdle muscular dystrophy 1B (LGMD1B), and dilated cardiomyopathy 1A (CMD1A) primarily affect muscle, whereas Dunnigan-type familiar partial lipodystrophy (FPLD) affects adipose tissue, and Charcot–Marie–Tooth neuropathy 2B1 (CMT2B1) impairs function of the peripheral nervous system (Rankin and Ellard, 2006; Worman and Bonne, 2007). A second class of laminopathies involves multiple tissues in different combinations, such as mandibuloacral dysplasia with type A lipodystrophy (MADA) and Hutchinson–Gilford progeria syndrome (HGPS; Rankin and Ellard, 2006; Worman and Bonne, 2007).

How mutations in LMNA cause disease and why many laminopathies are highly tissue specific are unclear and particularly intriguing, given that lamins A/C are expressed in most adult tissues (Rober et al., 1989). One model for the tissue specificity of laminopathies proposes that variants of lamins A/C may affect interactions with proteins that are themselves expressed in a tissue-specific manner (Schirmer and Gerace, 2005). In support, nuclear envelope proteomes are highly variable among tissues (Korfali et al., 2012). However, connecting altered lamin A/C protein interactions to clinically relevant molecular mechanisms remains a challenge.

To gain insights into the pathological mechanisms involved in laminopathies, we sought to systematically identify lamin A–interacting proteins whose binding is affected by disease-linked mutations. We screened an unbiased human ORFeome library comprising >12,000 genes to identify lamin A/C–binding proteins and probed their interaction with a comprehensive panel of 89 disease-linked lamin A variants. Our screening efforts identified novel lamin A/C-interacting proteins whose interaction is disrupted by laminopathy disease mutations, revealed lamin A/C residues important for protein–protein interactions, and identified disease-specific lamin A interactions at multidisease mutation sites. Our results support the notion that the tissue specificity of laminopathies is strongly driven by loss of tissue-specific lamin A/C interactions.

RESULTS

Systematic identification of lamin A–interacting proteins

To identify lamin A–interacting proteins in an unbiased manner, we performed a yeast two-hybrid (Y2H) screen using the human ORFeome V5.1 library, which contains 15,483 cloned open reading frames (ORFs), corresponding to 12,794 nonredundant human genes (a complete list of ORFs is available at http://horfdb.dfci.harvard.edu/hv5/). The library represents a comprehensive collection of ORFs and provides higher coverage than traditional Y2H libraries, the composition of which is limited by the gene expression pattern of the source tissue or cells from which the cDNA library is prepared, a potentially limiting factor when attempting to identify tissue-specific lamin A interactors (Bonaldo et al., 1996; Maier et al., 2011). The ORFeome library also ensures equal representation of each gene, in contrast to traditional Y2H libraries, in which the abundance of each clone in the library reflects its endogenous expression level. Because each ORFeome construct is sequence verified, the rate of false positives due to out-of-frame clones and mutations is minimized.

Each ORFeome clone was introduced into two Y2H expression vectors to generate a complete library of protein fusions between the ORF and Gal4 DNA–binding domain (DB-Y) and the Gal4 activation domain (AD-Y; Lamesch et al., 2007). Conversely, 41 unique lamin A fragments (Supplemental Table S1A), all of varying size and encoding different combinations of lamin A domains, were introduced into Y2H expression vectors to generate protein fusions with the Gal4 activation domain (AD-lamin A) or the Gal4 DNA–binding domain (DB-lamin A). The use of multiple lamin A fragments increases the throughput of candidate protein identification and ameliorates potential cytotoxic effects. DB and AD constructs were transformed into the haploid Saccharomyces cerevisiae strains Y8930 (MATα) and Y8800 (MATa), respectively, and two complementary screening strategies were performed (Figure 1A; Dreze et al., 2010). In a “forward” screen, 85 plates, each containing a different pool of 188 AD-ORFeome constructs, were tested against individual DB-lamin A constructs arrayed on a single plate (Figure 1A). In a “reverse” screen, the 41 AD-lamin A constructs were pooled and tested against individual DB-ORFeome constructs, which were arrayed across 147 plates (Figure 1A). Each screen was performed twice, and putative interactors were identified based on growth on selective media (see Materials and Methods; Dreze et al., 2010). As a control, yeast were replica plated onto media containing cycloheximide, which counterselects for the AD vector and aids in identifying DB-Y autoactivating proteins (Dreze et al., 2010). Approximately 1.2 million protein–protein combinations were tested, and positive yeast colonies were cultured and replated on selective and control media for secondary phenotyping. Only colonies that continued to grow on selective media, and not on control media, were considered primary hits. The identity of the primary hits was verified by yeast colony PCR amplification of the insert and end-read sequencing.

Overall 623 unique interactors were identified in the four screens (Figure 1A and Supplemental Table S2). As expected, the forward screen was less sensitive (∼50 hits/replica) compared with the reverse screen (∼420 hits/replica) due to decreased sampling, with more constructs screened per well in the forward screen (188 AD-ORFeome constructs vs. 41 AD-lamin A constructs). In addition, reporter sensitivity is lower in the forward screen because DB-lamin A acts as a transcriptional repressor when tethered to the GAL4 promoter (Lee et al., 2009). Of the 623 primary positives identified, 426 were retested in the Y2H assay with fresh archival stocks of the ORFeome library, resulting in 337 (79%) validated interactors (Figure 1, B and C, and Supplemental Table S1, C and D; see Materials and Methods). Four interactors (CCDC120, CMTM5, LMNA, and TRIM39) were identified in all four screens, and six interactors (AGTRAP, ARL6IP1, PIK3R2, SCARA3, STX4, and TOR1AIP1) were identified in three of the four screens and all validated in retests. We identified 333 interactors in two of four screens, and of 281 of these interactors 258 (∼92%) validated. Two hundred eighty interactors were identified in only one of four screens, and of 135 of these, 69 (∼51%) validated in retesting. Thirteen of the identified interactors, including LMNA, LMNB1, LMNB2, TMPO, and PCNA, had previously been reported as lamin A–interacting proteins, confirming the quality of the screening approach (Shumaker et al., 2008).

A subset of interactors was chosen randomly for biochemical validation in pull-down assays using N-terminally yellow fluorescent protein (YFP)-tagged interactors expressed in U2OS cell lines that stably express OneStrepTag-lamin A fusion (OST-lamin A) or OneStrepTag alone (OST; Figure 2; Kubben et al., 2010). Twelve of 18 Y2H interactors were detectable in pull-down assays, which included known lamin A interactors such as LMNA itself, TMPO, and LMNB1, in addition to several of the novel lamin A interactors (Figure 2). These results demonstrate that the interaction data set is enriched for lamin A–interacting proteins.

Cellular localization of interactors

Given that lamin A accumulates prominently at the nuclear periphery, we tested the localization of a subset of 104 interactors by high-throughout microscopy after expression in U2OS cells of N-terminally YFP-tagged versions of the interactors (Figure 3 and Supplemental Table S2). The localization pattern of each candidate was visually scored and categorized according to the extent of NE accumulation and cellular localization (Figure 3A). Several known lamin A interactors were included in the visual screen, such as LMNA, LMNB1, LMNB2, and TMPO, which served as positive controls for NE accumulation, and untagged YFP alone, which served as a negative control. We found that 53% of interactors showed NE accumulation (Figure 3B) and 47% of interactors did not accumulate at the NE but were homogeneously distributed throughout the nucleus and/or cytoplasm (Figure 3B). Note that peripheral localization is not a prerequisite for lamin interactors, as lamin A is also present in the nuclear interior, and several bona fide lamin interactors are diffusely distributed throughout the nucleus (Hozak et al., 1995; Dorner et al., 2007). Of the interactors that accumulated at the NE, 17% were exclusively nuclear, whereas the remaining candidates localized to the nucleus and cytoplasm. Of the interactors that did not accumulate at the NE, only 8% were exclusively localized in the cytoplasm. (Figure 3B).

Identification of progerin-specific interactors

To apply the ORFeome screening strategy to a specific laminopathy, we analyzed the Y2H interactome of progerin, the lamin A isoform responsible for the premature aging syndrome HGPS (Eriksson et al., 2003). Progerin is a dominant lamin A isoform generated by the G608G mutation, which activates an intronic alternative pre-mRNA splice site (Eriksson et al., 2003). Owing to the activation of the internal splice site, progerin has a 50–amino acid deletion in the C-terminal tail, which leads to its permanent farnesylation (Glynn and Glover, 2005). Five progerin constructs of varying length were tested in the primary screen, and the resulting validated candidates were classified according to their interaction with lamin A and/or progerin (Supplemental Table S1C; see Materials and Methods). We identified 225 progerin-specific interactions, 51 proteins that interacted with both lamin A and progerin (Supplemental Table S1D), and 61 interactors that were specific to lamin A. Of 80 progerin-specific interactors tested for their cellular localization, 89% displayed NE staining (Figure 3C), compared with only 53% of lamin A (Figure 3B). Of interest, only 8% of progerin-specific interactors, which accumulated at the NE, were also localized exclusively in the nucleus, in contrast to 17% of lamin A–specific or lamin A/progerin interactors (Figure 3, B and C).

Gene Ontology (GO) analysis confirmed the notion that progerin-specific interactors are a distinct group of proteins from lamin A interactors. Whereas lamin A interactors were enriched for components of the NE and lamina (Figure 4A; p < 1 × 10⁻³), progerin-specific interactors comprised mostly integral and intrinsic membrane proteins and components of the endoplasmic reticulum (ER; Figure 4B; p < 1 × 10⁻³). The progerin-specific interactors were enriched for proteins with soluble N-ethylmaleimide–sensitive factor attachment protein receptor activity and channel and transmembrane transporter activity (Figure 4B; p < 1 × 10⁻³). Progerin-specific interactors were also enriched for proteins that function in vesicle-mediated transport, membrane organization, transmembrane transport, and establishment of protein localization (Figure 4B; p < 1 × 10⁻³). The higher percentage of progerin-specific interactors that displayed NE staining compared with lamin A interactors might be explained by the preponderance of membrane-related components in this group, as high levels of membrane proteins might aggregate in the ER, which is continuous with the NE.

The observation that a higher percentage of progerin-specific interactors are not exclusively found in the nucleus but are also in the cytoplasm suggests that interactions may reflect pre–lamin A processing, which is thought to take place, in part, in the cytoplasm (Barrowman et al., 2008). A key step in the maturation of lamin A is the addition and eventual removal of a C-terminal farnesyl group (Weber et al., 1989), whereas progerin is permanently farnesylated (Glynn and Glover, 2005). To test for a potential role of farnesylation in the identified interactions, we probed progerin interactors against several lamin A variants impaired in farnesyl processing (Table 1). Of 223 progerin-specific candidates tested, none maintained its interaction with a progerin fragment containing a C661S mutation, which blocks transfer of the farnesyl moiety onto the terminal cysteine in lamin A or progerin (Table 1; Capell et al., 2005). The effect of the C661S variant was specific to the progerin-only interactors and did not globally disrupt all interactions with progerin, since of the 27 candidates that interacted both with progerin and lamin A, 9 (33%) maintained their interaction with progerin C661S (Table 1). Conversely, of 220 progerin-specific interactors tested, 165 (75%) gained interaction with lamin A L647R, a variant that causes permanent farnesylation by blocking cleavage of the farnesyl group by the lamin A endoprotease ZMPSTE24 (Pendas et al., 2002; Corrigan et al., 2005; Glynn and Glover, 2005). These observations demonstrate that the majority of progerin interactions are mediated by its farnesylated C-terminus and that protein farnesylation is sufficient to mediate lamin A–protein interactions.

Systematic identification of interactors involved in laminopathies

Next we sought to identify disease-relevant interactors more broadly by testing a subset of 58 interactors, chosen based on their ability to interact robustly with multiple lamin A fragments, against a panel of 89 lamin A variants that represent 15 known laminopathies and are distributed throughout the lamin A protein (Figure 5 and Supplemental Table S3). We used the Y2H assay to test 4918 unique interactor–lamin A mutant binary pairs and identified 554 pairings that resulted in a loss of interaction (Figure 5 and Supplemental Table S3). The number of lost interactions among lamin A mutants ranged from 0 (0/42) to 75% (43/57; Figure 6; see later discussion).

Four of the lamin A mutants did not lose interaction with any of their binding partners. One of them, Arg377His, is located within the α-helical coiled coil 2B domain and is implicated in limb girdle muscular dystrophy D1, Emery–Dreifuss muscular dystrophy 2, and dilated cardiomyopathy 1A, whereas the remaining three, Asp136His, Glu145Lys, and Glu159Lys, are located within the α-helical coiled-coil 1B domain and are associated with HGPS (Muchir et al., 2000; Eriksson et al., 2003; Sebillon et al., 2003; Reichart et al., 2004; Garg et al., 2009). Together with Ser143Phe, a fourth HGPS-associated lamin A variant, HGPS variants within the coiled-coil 1B domain had only a minimal effect, losing interaction with 2.5% of candidates compared with 17% for the rest of the variants in the coiled-coil 1B domain. The tendency of HGPS-associated variants to have little or no effect on protein–protein interactions supports a model in which disturbed assembly of lamin filaments, which leads to dominant gain-of-function effects, is the cause of HGPS (Goldman et al., 2004; Dahl et al., 2006). The effects of the HGPS-associated variants probably manifest as perturbations in higher-order filament formation, since none of the variants prevented lamin dimer formation (Supplemental Table S3).

Structure–function interaction analysis

The availability of a comprehensive library of lamin A mutants allowed mapping of the lamin A protein domains required for interactions. Of note, lamin A variants inside the immunoglobulin (Ig)-like fold domain lost interaction with more candidates than variants in the unstructured C-terminal tail (Table 2). The Ig-like fold domain is composed of two β-sheets that are closely associated in a β-sandwich (Dhe-Paganon et al., 2002; Krimm et al., 2002). The side chains of some residues orient toward the core of the domain and participate in stabilization of the β-sandwich through hydrophobic contacts and hydrogen bonds (Dhe-Paganon et al., 2002; Krimm et al., 2002). As expected, the most severely affected Ig-like domain mutants are two core variants, Leu454Pro and Asn456Lys, which lost interaction with 72 and 75% of candidates tested, respectively (Table 3). The third most severely affected Ig-like domain mutant was Arg455Pro, which lost interaction with 40% of candidates (Table 3). Unlike the side chains of Leu454 and Asn456, which are both positioned within the core of the Ig-like fold, a structural role of Arg455 has not been determined. The large effect of the Arg455Pro variant suggests that like Leu454 and Asn456, the side chain of Arg455 is oriented within the core and important for the stability of the Ig-like fold domain. Asn456Ile lost interaction with 18% of candidates, whereas Asn456Asp lost interaction with only 1% of candidates. The different effects for various Asn456 variants are likely a reflection of how the core domain accommodates the structural differences between amino acid side chains.

Side chains situated on the surface of the Ig-like fold domain, such as Trp482, are not predicted to affect the overall stability of the Ig-like fold domain, but are expected to affect protein–protein interactions (Dhe-Paganon et al., 2002). Three different Trp482 variants, Trp482Gln, Trp482Leu, and Trp482Trp, lost interaction with 4% of candidates (Table 3), which is comparable to the 6% average of all solvent-exposed residues (Table 2). Overall lamin A variants whose substitutions affect core residues involved in maintaining the β-structure lost interactions with on average 20% of the candidates, compared with 6% for 8 variants affecting solvent-exposed residues and 9% for 15 variants affecting residues without a determined orientation, in line with a critical global role of the β-sheet structure in mediating interactions of lamin A with its partners (Table 2).

The relationship between the number of lost interactions and the location of lamin A mutations within the Ig-like domain fits a previously noted correlation of disease type and location of lamin A mutations within the Ig-like domain (Dhe-Paganon et al., 2002). Our study included 23 variants that are predicted to affect the Ig-like domain core. Of these 21 are implicated in EDMD2, LGMD1B, CMD1A, and other types of muscular dystrophies and lose interaction with 20% of candidates. In comparison, of the eight variants in our study predicted to have surface-exposed residues, six cause FPLD2, HGPS, and MADA and lose interaction with only 6% of candidates. These results agree with the noted correlation between Ig-like fold structure and resulting diseases. Ig-like fold domain mutations associated with muscular dystrophies and cardiomyopathies generally affect core residues and disrupt overall protein stability, whereas mutations associated with lipodystrophies and progeria are surface-exposed residues and do not affect overall domain stability.

Interactors affected by laminopathy variants

In a converse approach, we asked how many interactions each lamin A interactor lost when tested against the panel of 83 lamin A mutants (Table 4). Of the 58 interactors tested, 50 lost interaction with at least one mutant (Table 4 and Figure 7). The interaction of eight candidates with lamin A was insensitive to lamin A mutations. Of importance, this group includes lamins LMNA, LMNB1, and LMNB2, suggesting that none of the variants significantly impairs lamin dimer formation mediated through the central coiled-coil domain. The other lamin A interactions that were insensitive to mutation were proliferating cell nuclear antigen (PCNA), reticulon 1/3 (RTN1/3), CREB3, and SENP2, suggesting that these proteins contain multiple, redundant lamin A interaction sites or that the relevant lamin A mutants that affect their interaction is not present in the panel. For the 50 interactors that lost binding with lamin A mutants, the number per candidate ranged from 1 to 72% (Table 4 and Figure 7). Of note are the combined losses of TMPO, TOR1AIP1, and DUSP13, which shared 10 mutant variants with which they lost interaction. Remarkably, 9 of the 10 variants cause CMD1A or EDMD2, which are diagnosed as separate diseases but have overlapping symptoms. Mutations in TMPO are known to cause CMDT1A, a cardiomyopathy, whereas DUSP13 and TOR1AIP1 have not previously been implicated in any diseases. All three interactors likely play a role, given the significant overlap among the candidates, the known association of TMPO with cardiomyopathies, and the similarities between EDMD2 and cardiomyopathies.

Disease-specific interactions at multidisease mutation sites

A remarkable feature of laminopathies is that phenotypically vastly distinct, tissue-specific diseases can be caused by differential amino acid substitutions of the same residue. A prominent example is Thr-528. Patients with Thr528Lys or Thr528Arg mutations suffer from muscular dystrophies, whereas Thr528Met causes progeria and a lipodystrophy as compound heterozygote with different lamin A variants (Bonne et al., 2000; Raffaele Di Barletta et al., 2000; Vytopil et al., 2003; Savage et al., 2004; Verstraeten et al., 2006; Benedetti et al., 2007; Yuan et al., 2010; Scharner et al., 2011; Saj et al., 2012). To test whether emergence of distinct diseases is related to loss of interaction with distinct sets of proteins, we tested lamin A isoforms with the three Thr-528 substitutions against a panel of 57 interactors. We identified 35 interactors that lost binding with at least one of the Thr-528 variants (Supplemental Table S3). Remarkably, the candidates largely grouped by disease (Figure 8). Nineteen of the interactors lost association with the variants causing muscular dystrophies—Thr528Arg and Thr528Lys—but their interaction with Thr528Met, which causes HGPS and a lipodystrophy, was not affected. Conversely, 11 proteins lost interaction with the Thr528Met variant but not the Thr528Arg or Thr528Lys variants. The differential interaction properties correlate with the steric orientation of the mutant amino acids. The side chain of Thr-528 forms part of the core of the Ig-like fold domain, and substitutions, particularly with positively charged amino acids such as Arg and Lys, are predicted to disrupt folding and possibly reduce protein stability. In contrast, methionine, which is neutral and hydrophobic, more closely resembles threonine and is less likely to cause major structural changes.

Additional multidisease mutations exist in LMNA. Patients with Arg133Pro mutations develop EDMD2, and those with Arg133Leu mutations develop Werner syndrome (WS) or lipodystrophies. We identified two interactors, DUSP13 and TMPO, that specifically lost interaction with Arg133Pro but not Arg133Leu. Another multidisease mutation site is Arg541. Arg541His and Arg541Pro mutations cause EDMD2, whereas Arg541Cys, Arg541Ser and Arg541Gly primarily cause CMD1A. We identified one candidate, MTHFD2, that lost interaction with Arg541 variants associated with EDMD2 but maintained interaction with variants associated with CMD1A. We conclude that variants at multidisease sites that cause related diseases show specificity for interaction partners, possibly explaining the tissue specificity of their appearance.

DISCUSSION

Mutations in lamins and other proteins of the nuclear lamina have been linked to diverse pathologies ranging from muscular dystrophies to lipodystrophies, neuropathies, and progeroid aging syndromes (Rankin and Ellard, 2006; Worman and Bonne, 2007). Although lamins A/C are widely expressed in most adult tissues, many of these diseases exhibit pronounced tissue specificity. It has been proposed that disrupted lamin A interactions with unidentified tissue-specific factors may mediate the development and tissue preferences observed in these diseases (Schirmer and Gerace, 2005). This model is in line with the observed high degree of tissue specificity of the NE proteome (Korfali et al., 2012). To unravel the pathogenic mechanisms in laminopathies, we screened an unbiased ORFeome library to identify lamin A interactors and systematically evaluated their interaction with 89 different lamin A disease variants.

We identified and validated 337 proteins that interact with lamin A or progerin. Reassuringly, the lamin A interactors included many known interactors, were significantly enriched for components of the lamina, NE, and NE membrane, and preferentially localized to the nuclear periphery. The high validation rate in orthogonal pull-down assays suggests that our set of candidates contains a significant number of genuine lamin A interactors. Comparison of the Y2H candidates with available NE proteomic data sets obtained from muscle and liver tissues and blood leukocytes indicated that of the 337 candidates, 35% are also found in NE biochemical fractions (Korfali et al., 2012). It is important to note that the characterization of the NE proteome is incomplete and is variable across different tissue and cell types. Only 16% of NE proteins overlap between muscle, liver, and leukocytes, suggesting that the actual overlap between our candidates and the NE proteome is higher (Korfali et al., 2012).

We identified 102 lamin A interactors, several of which are associated with at least one of the laminopathies or related diseases. These include LMNB2, which is associated with an acquired lipodystrophy, and TMPO, 2-actin, phospholamban, transmembrane protein 43 (TMEM43/LUMA), and vesicle-associated membrane protein 1/synaptobrevin 1, which are implicated in cardiomyopathies and other heart diseases. Whereas LMNB2 maintained its interaction with all lamin A variants, TMPO lost interaction with 11 variants, 9 of which cause cardiomyopathies or muscular dystrophies. It will be important to investigate the involvement of TMPO in cardiomyopathies caused by mutations in LMNA, particularly those identified here.

In addition, our approach has the potential to identify novel disease-relevant candidate proteins in an unbiased manner. One lamin A interactor, annexin VI (ANXA6), lost interaction with eight variants located within the Ig-like domain. Seven of the variants are buried within the core and cause EDMD2, whereas the orientation of the remaining variant is unknown and causes CMD1A. These findings dovetail with indications that ANXA6 may have a role in cardiomyopathies. Annexins are Ca²⁺ effectors that mediate cellular responses to changing intracellular levels of calcium, and elevated intracellular calcium concentrations cause the translocation of several annexins, including ANXA6, to the plasma membrane and nuclear envelope (Skrahina et al., 2008). Proteomic studies show that ANXA6 is a component of the NE in muscle, liver tissue, and blood cells (Korfali et al., 2012). Overexpression of ANXA6 in cardiomyocytes of a transgenic mouse model causes hypertrophy and heart failure (Gunteski-Hamblin et al., 1996). During end-stage heart failure, levels of ANXA6 decrease in cardiomyocytes, whereas other annexins show increased levels (Jans et al., 1998). This has led to the view that down-regulation of ANXA6 during heart failure creates a framework that favors improved cardiomyoctye function (Gunteski-Hamblin et al., 1996). The involvement of ANXA6 in cardiomyopathies and muscular dystrophies warrants further study.

We tested 89 lamin A variants against 58 of the candidates that interact with lamin A, and we generated a comprehensive map of effects of variants on protein–protein interactions. This data set allowed us to understand which residues and protein domains are important for protein–protein interactions. Approximately half of the N-terminal portion of lamin A protein consists of coiled-coil domains. Mutations within the coiled-coil domains, including the linkers, lost interaction with on average 10% of candidates. The coiled-coil domain mediates dimerization with other lamins, but none of the lamins lost interaction with any of the variants, indicating that none of the variants participates in dimerization. The C-terminal half of lamin A, excluding the Ig-like domain, is unstructured and lost interaction with on average ∼6% of candidates. Except for a nuclear localization signal, the function of the unstructured domain is not well understood. Structural studies of the Ig-like domain indicate that residues whose side chains are oriented toward the core are crucial for stabilization of the β-sandwich (Dhe-Paganon et al., 2002; Krimm et al., 2002). Consistent with the structural predictions, core variants lost interactions with on average 20% of candidates, compared with 6% of candidates for variants on the domain surface.

Systematic screening of known disease mutations has the potential to provide novel insights into disease mechanisms. The identification of a disproportionate number of progerin-specific interactors was unexpected, since only 5 of 41 lamin A constructs used in the primary screen contain the disease-relevant 50–amino acid deletion. The identification of gain-of-function progerin interactors is consistent with models of HGPS, which envision a dominant effect of progerin (Goldman et al., 2004; Dahl et al., 2006). Of interest, the 225 progerin-specific interactors were significantly enriched for membrane proteins, possibly reflecting a contribution of the C-terminal farnesyl group to their interactions. Conversely, 61 of 112 (54%) lamin A interactors lost interaction with progerin. Of the eight candidates that did not lose interaction with any of the missense lamin A variants, five lost interaction with progerin, including CREB3, PCNA, LMNB2, and two reticulons, RTN1 and RTN3. RTN4, a third reticulon, and ARL6IP1, a candidate that also contains the reticulon homology domain (RHD), also lost interaction with progerin. In variant tests with lamin A missense mutations, RTN4 and ARL6IP1 lost interactions with only 2 and 3% of candidates, respectively (Figure 7), suggesting that the proteins containing the RHD are largely insensitive to laminopathy missense mutations but are sensitive to the deletion in progerin. Reticulons and ARL6IP1 have membrane-shaping activity, function as ER-shaping proteins (Yang and Strittmatter, 2007; Yamamoto et al., 2014), and also function in NE assembly by regulating the contribution of ER tubules to the nascent NE and in subsequent NE expansion (Anderson and Hetzer, 2008). Nuclei from HGPS patients are lobulated and misshapen, which may indicate dysfunctional NE reformation and expansion.

Why different substitutions of the same lamin A residue cause vastly different diseases is an enigmatic feature of the laminopathies. In our Y2H analyses, multimutation residues that cause distinct diseases, such as Arg-133, Thr-528, and Arg-541, led to disease-specific loss of distinct sets of interactors. The results substantiate the genetics of laminopathy patients with molecular evidence, demonstrating that different substitutions of the same residue can have profoundly different effects on lamin A protein function. Identification of disease-specific candidates is a foothold for future functional studies and clinical efforts.

Lamin A/C is at the crossroads of many nuclear processes, and mutations cause a large number of diverse diseases. It is becoming increasing clear that the molecular functions of lamin A/C go far beyond its traditional role as a structural element and involve an extensive number of lamin A/C binding partners. Systematic identification of these interaction partners will provide the basis for the functional analysis of lamin A and its disease-causing mutants.

MATERIALS AND METHODS

Cloning, yeast transformations, and pool preparations

Fragments of lamin A were amplified from cDNA using PfuUltra II Fusion HS DNA Polymerase (Agilent Technologies, Santa Clara, CA), cloned into the pDONR223 (Life Technologies, Carlsbad, CA) vector, and then verified by sequencing. See Supplemental Table S1A for a list of all lamin A fragments. pDONR223-lamin A fragments were then cloned into pAD/pDB vectors using LR Clonase II Plus enzyme (Life Technologies). Yeast strains Y8800 and Y8930 were transformed with pDB-lamin A and pAD-lamin A vectors, respectively, using a lithium acetate transformation protocol (Dreze et al., 2010). An AD pool of 41 different pAD-lamin A fragment constructs transformed in Y8800 was constructed as described in Dreze et al. (2010).

Yeast two-hybrid screen

Screening was performed essentially as described in Dreze et al. (2010). Before the screen, the AD-/DB-lamin A constructs were tested to ensure that none of the fusion proteins were autoactivators. Glycerol yeast stocks of the DB-X (Y8930) or AD-pool (Y8800) strains were used to inoculate 96-well plates containing liquid selective media (SC/-Leu or SC/-Trp). Plates were incubated at 30ºC for 3 d with agitation. For the mating reaction, 5 μl of opposite-mating-type strains was combined in 96-well plates containing 160 μl of liquid yeast extract/peptone/dextrose (YEPD) per well and incubated overnight at 30ºC with shaking. A 5-μl amount of each mating reaction was replica plated onto SC/-Trp/-Leu/-His + 1 mM 3-amino-1,2,4-triazole (3-AT) media (screening plate) and SC/-Leu/-His + 1 mM 3-AT + 1 mg/l cycloheximide media (de novo autoactivator plate). Each set of plates was incubated at 30ºC for 3 d and scored for primary positives. Primary positives were defined as having a growth phenotype on the screening plate and a negative growth phenotype on the corresponding de novo autoactivator plate. Primary positive colonies were picked from the screening plate with sterile toothpicks and used to inoculate 96-well plates containing 160 μl of liquid SC/-Leu/-Trp media per well and then grown for 2 d at 30ºC. Then 5 μl of culture from the SC/-Leu/-Trp plate was spotted onto fresh selective agar medium plates and de novo activator plates to verify the primary positive phenotype (phenotyping II). After 3 d, plates were again scored for secondary positives, which were then used to inoculate 96-well plates containing 160 μl of fresh liquid SC/-Leu/-Trp media per well and grown for 2 d at 30ºC. Sequence identification of primary positives was performed by yeast colony PCR, followed by sequencing of the PCR product. To reduce false-positive rate and improve the quality of the interaction data set, primary positives were retested in the Y2H assay using fresh archival stocks of the ORFeome library. This archival library was prepared independently from the library used in the primary screen and helps to reduce the number of false positives caused by cis- or trans-acting mutations. Of the 623 primary positives in the primary screen, 426 (68%) were present in the retest. These primary positives were then tested against lamin A using a pooling strategy similar to that used in the primary screen, except that the 41 lamin A constructs were divided into six subpools in which each pool was assigned 5–11 lamin A constructs (note that 7 additional lamin A constructs not in the primary screen were included in the retest; see Supplemental Table S1, C and D, for composition of each pool and retest results). The six AD/DB-lamin A subpools were individually tested against the 426 AD/DB-ORFeome primary positive candidates (Figure 1B). If the candidate scored positive with at least one of the six lamin A subpools, it was considered verified.

Transfections and pull-down assay

U2OS cells stably expressing OneSTrEP-tagged lamin A (OST-lamin A) or the OneSTrEP epitope alone (OST; IBA, Goettingen, Germany) were grown until ∼90% confluence in 150-mm plates. Cells were transfected with YFP-candidate fusion constructs (pEYFP candidate) using GenJet reagent according to the manufacturer's instructions. After 24 h, cells were subjected to the pull-down procedure as previously described (Kubben et al., 2010). We added 2× NuPAGE loading buffer (Life Technologies) to washed STrEPtactin matrix (IBA, Goettingen, Germany) and then heated it at 95˚C for 20 min to elute precipitated proteins.

GO analysis

GO analysis of interactors was performed for GO categories cellular components, biological process, and molecular function using ORFeome v5.1 (http://horfdb.dfci.harvard.edu/hv5/) as background (Huang da et al., 2009). Entrez Gene IDs were used as input, and the default Database for Annotation, Visualization, and Integrated Discovery settings were used for analysis.

Western blot

Samples were resolved on NuPAGE Bis-Tris Precast 4–12% gels (Life Technologies) and then blotted on polyvinylidene fluoride membranes (EMD Millipore, Billerica, MA). Membranes were blocked for 1 h in blocking buffer (5% bovine serum albumin, 20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Tween-20) and then incubated overnight at 4°C with anti-GFP antibody (Life Technologies) diluted 1:2000 with blocking buffer. The membranes were washed three times for 15 min with TBS-T (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Tween-20) and incubated with horseradish peroxidase–labeled anti-mouse (ECL mouse IgG; GE Healthcare, Little Chalfont, United Kingdom) diluted 1:5000 with blocking buffer for 1 h at room temperature and washed as before. Proteins were detected with Amersham ECL Plus Western Blotting reagent (GE Healthcare). Aliquots of pre–pull-down cell lysate for each sample were analyzed for levels of β-tubulin to ensure uniform protein levels.

Localization of candidates

The pEYFP-Y candidate constructs used in the visual screen were made by using available pAD/pDB-Y clones from the yeast ORFeome library. Yeast minipreps were performed using Zymoprep Yeast Plasmid Miniprep II (Zymo Research, Irvine, CA) according to the manufacturer's instructions. Rescued plasmids were transformed into TOP10-competent cells (Life Technologies), and candidate ORFs were amplified using miniprep DNA as template with PfuUltra II Fusion HS DNA Polymerase, using the forward primers AD (5′-CGCGTTTGGAATCACTACAGGG-3′) and DB (5′-GGCTTCAGTGGAGACTGATATGCCTC-3′), and the reverse primer Term (5′-GGAGACTTGACCAAACCTCTGGCG-3′). PCR products were recombined into a Gateway compatible pEYFP vector using LR Clonase II Plus enzyme (Life Technologies). Sequence-verified pEYFP constructs were transfected using GenJet In Vitro DNA Transfection Reagent (SignaGen Laboratories, Rockville, MD) into U2OS cells in 96-well clear-bottom microplates according to the manufacturer's instructions. Cells were fixed 48 h posttransfection with 4% paraformaldehyde in phosphate-buffered saline and stained with DRAQ5 (Cell Signaling Technology, Danvers, MA). Confocal fluorescence images of multiple fields from each well were captured using the CellVoyager CV6000 high-content microscopy system (Yokogawa Electric Corporation, Tokyo, Japan) and analyzed using ImageJ software (National Institutes of Health, Bethesda, MD).

Y2H lamin A–variant panel construction

A series of laminopathy variants was created in three fragments of lamin A: lamin A 2–664, lamin A 259–664, and lamin A 388–664. Mutations were introduced into fusion PCR products by site-directed mutagenesis (for a complete list of primers and variants see Supplemental Table S4), which were subsequently recombined into pDONR223 using BP Clonase and then verified by sequencing (Suzuki et al., 2005; Charloteaux et al., 2011). This resulted in three variant panels: 1) pDONR223-lamin A 2–664 (n = 88 variants), 2) pDONR223-lamin A 259–664 (n = 73 variants), and 3) pDONR223-lamin A 388–664 (n = 62 variants). Collectively the three panels comprise 223 unique pDONR223-lamin A constructs and represent 89 unique laminopathy variants. The pDONR223-lamin A construct was then recombined into the Y2H vectors pAD and/or pDB using LR Clonase. This resulted in five variant panels: 1) pAD-lamin A 2–664 (n = 83 variants), 2) pAD-lamin A 259–664 (n = 59 variants), 3) pDB-lamin A 2–664 (n = 73 variants), 4) pDB-lamin A 259–664 (n = 73 variants), and 5) pDB-Lamin A 388–664 (n = 62 variants). All pAD-lamin A and pDB-lamin A constructs were used to transform haploid yeast strains Y8800 and Y8930, respectively, and are maintained as glycerol stocks.

FOOTNOTES

ACKNOWLEDGMENTS

REFERENCES