Capturing the Conversion of the Pathogenic Alpha-1-Antitrypsin Fold by ATF6 Enhanced Proteostasis

Abstract

Genetic variation in alpha-1 antitrypsin (AAT) causes AAT deficiency (AATD) through liver aggregation-associated gain-of-toxic pathology and/or insufficient AAT activity in the lung manifesting as chronic obstructive pulmonary disease (COPD). Here, we utilize 71 AATD-associated variants as input through Gaussian process (GP) based machine learning to study the correction of AAT folding and function at a residue-by-residue level by pharmacological activation of the ATF6 arm of the unfolded protein response (UPR). We show that ATF6 activators increase AAT neutrophil elastase (NE) inhibitory activity, while reducing polymer accumulation for the majority of AATD variants, including the prominent Z-variant. GP-based profiling of the residue-by-residue response to ATF6 activators captures an unexpected role of the ‘gate’ area in managing AAT specific activity. Our work establishes a new spatial covariant (SCV) understanding of the convertible state of the protein fold in response to genetic perturbation and active environmental management by proteostasis enhancement for precision medicine.

eTOC Blurb:

Using genetic variation in the human population through Gaussian process (GP) based machine learning, Sun et al. capture the residue-by-residue conversion of the pathogenic alpha-1 antitrypsin (AAT) fold into a functional protein with improved activity and reduced polymer by pharmacological activation of the ATF6 branch of unfolded protein response (UPR).

Graphical Abstract

Introduction

About 30% of the human proteome is inserted into the endoplasmic reticulum (ER) by the ribosome where it can be N-glycosylated, folded and trafficked through the secretory pathway for function in downstream organelles, at the cell surface, or in the extracellular space ^1–3. Protein folding/trafficking versus degradation decisions in the ER are regulated by the stoichiometries of proteostasis network components ^4–13, which are controlled by the unfolded protein response (UPR) and to a lesser extent by other stress-responsive signaling pathways ^2,3,14,15. These components include chaperones and folding enzymes, proteins that control trafficking, components of the ubiquitin proteasome system (UPS) responsible for ER-associated degradation (ERAD), and proteins involved in the endosomal-lysosomal degradation systems ¹⁶. Our understanding of the role(s) of the proteostasis machinery in the management of protein folding in response to genetic variation in the population to generate function (the genotype-to-phenotype transformation) driving human health and disease is poor.

Inherited genetic variants that challenge ER protein folding can lead to excessive protein degradation and/or aggregation implicated in etiologically diverse protein misfolding diseases ^{4–13,17–24}. One such disease is alpha-1 antitrypsin (AAT) deficiency (AATD). Alpha-1 antitrypsin (AAT) deficiency (AATD) is a complex aging-linked inherited genetic disease of the liver-to-lung axis ^25–29. AATD is caused by mutations in the SERPINA1 gene that can cause aggregation-associated gain-of-toxic pathology in the liver and/or loss-of-AAT function in the lung leading to chronic obstructive pulmonary disease (COPD). Wild-type (WT) AAT is primarily synthesized and secreted by liver hepatocytes at a rate of 2 grams per day. This rate can increase 3 to 4-fold in response to a variety of stressors including tissue injury and inflammation. As the most abundant serine protease inhibitor secreted into the plasma for transfer to the interstitial space of the lung, AAT plays a critical role in preventing the degradation of lung tissue by inhibition of neutrophil elastase (NE) protease activity. Excessive NE activity can lead to COPD, consistent with its primary role as a modifier of lung inflammatory disease based on GWAS ^25,30–32. NE inhibition occurs when the reactive central loop (RCL) of the conformationally constrained AAT molecule is attacked by NE to form a covalent conjugate ³³. The cleaved portion of RCL moves the bound NE from one end of the AAT structure to the other end by inserting itself as a new β-strand 4 in the central β-sheet A of AAT ^33–38. This ‘loop-sheet’ insertion mechanism distorts the active site of NE to form a neutralized AAT-NE conjugate that is degraded through the endocytic-lysosomal pathways ^36,39.

The vast majority of severe AATD cases (~95%) are caused by AAT-Z allele (Glu366Lys; residue numbering includes signal peptide sequence), which affects about 1 in 1600–3000 live births in people of European descent. Inheritance of AAT-Z leads to misfolding and polymerization of AAT in the ER of hepatocytes ^25,40–46, which can trigger early onset neonatal hepatitis and juvenile cirrhosis in approximately 10% of AATD population ^47,48. Furthermore, ~35% of adult AATD patients showed clinically significant liver fibrosis suggesting a slowly progressing AAT aggregation-associated gain-of-toxic pathology in liver hepatocytes ^49,50. A secretion deficiency of functional AAT-Z into plasma can lead to COPD that begins to present at 30–50 years of age, and progression can require a lung and/or liver transplant ²⁵. Augmentation therapy by infusing of WT AAT is the current standard of care - an approach that has minimal impact on the progression of liver or lung disease ^51,52, possibly because extracellular AAT aggregates also contribute to COPD pathology. While small molecule pharmacological chaperones have been reported to enhance AAT-Z secretion and reduce intracellular AAT-Z aggregation, the chemotypes discovered to date stabilize an AAT conformation that cannot react with and inhibit NE ^41,53,54. Thus, there is no reason to believe that these chemotypes would address the deficiency of NE inhibition in COPD. Besides the AAT-Z mutation, there are >600 variants reported in SERPINA1 gene in the Genome Aggregation Database (gnomAD) ⁵⁵, including 277 missense variants, ~100 of which are associated with clinical disease ^28,56,57. The different variants in the population lead to diverse AAT folding and functional challenges to successfully develop strategies that can broadly intervene AATD population.

One potential strategy to broadly intervene in the genetic and phenotypic diversities found in AATD is the activation of adaptive signaling pathways that manage multiple proteostasis pathways. The regulation of the composition of ER proteostasis network is managed by the three signaling arms of the UPR comprising the activating transcription factor 6 (ATF6) arm, the inositol-requiring enzyme 1α-X box binding protein 1s (IRE1-XBP1s) arm and the protein kinase regulator (PKR)-like ER kinase (PERK) arm ^2,3,14,15. Intriguingly, the ATF6 arm of the UPR promotes adaptive remodeling of ER proteostasis pathways that improve ER proteostasis of multiple disease-associated proteins ^2,17,58–68. ATF6 is a transmembrane leucine zipper transcription factor that is cleaved by site-1 protease (S1P) and site-II protease (S2P) to generate the ATF6 N-terminal fragment that can homodimerize to induce an adaptive transcriptional program that alters the composition of ER proteostasis network components ¹⁷. Given the central role of proteostasis management in biology ^{10,13,16,69,70}, application of arm-selective small molecule UPR activators that can reprogram the ER proteostasis network ^{2,3,12,17,62,64,71–74} offer the potential to correct mutant protein folding, trafficking and function to ameliorate proteinopathies such as AATD.

To mechanistically capture the response of protein fold and function to therapeutics, we have recently developed a Gaussian process (GP) based machine learning approach, called Variation Spatial Profiling (VSP), that uses genetic variation in the population and the associated phenotypes as input to generate spatial covariance (SCV) relationships for each residue in the context of all other residues of a protein sequence to define the molecular mechanism(s) facilitating normal function and their response to pharmacological intervention as output ^75–81. Using VSP to analyze the differential impact of 71 AAT variants found in the AATD population to understand the response of all residues in wild-type AAT in triggering disease, we have previously shown distinct clusters of residue-residue SCV relationships in the AAT protein sequence that are differentially responsible for NE inhibitory activity, monomer secretion and/or polymer formation ⁸⁰. How these residue-residue SCV relationships respond to UPR mediated reprogramming of the ER proteostasis network to manage the folding and function of AAT remains unknown.

Here, we used the GP-based VSP analysis to define how selective pharmacological ATF6 activators such as AA-147 ^59,64 influence the folding, trafficking, aggregation, and activity of disease-associated AATD variants. Strikingly, we find that the reprogramming of ER proteostasis network capacity through pharmacological ATF6 activation can remodel residue-residue SCV relationships to increase NE inhibitory specific activity, while reducing the intracellular polymer accumulation across the majority of disease-associated AATD variants. Through GP based analyses it is now clear that pharmacological ATF6 activation has noteworthy potential as a correction strategy for the global AATD population with aggregation-associated liver disease and/or loss-of-function associated COPD. These results demonstrate the potential of pharmacological proteostasis network reprogramming as a means to manage different challenges associated with mutant AAT folding and function. In general, GP analysis of the impact of proteostasis on natural variation in the population provides a precision framework to understand and modulate the convertible state of protein fold and its function at atomic resolution in a broad range of aggregation diseases affecting human biology.

Results

ATF6 activators correct the trafficking, conformation, secretion and function of AAT-Z

To address the impact of pharmacological ATF6 activation on AATD, we first measured the influence of two structurally distinct ATF6 activators AA-147 and AA-263 ^59,61 on the severe pathogenic mutation AAT-Z (Fig. 1A; used at 10 μM). Their effect on endoplasmic reticulum (ER) to Golgi trafficking and total secretion of AAT-Z was measured using transiently transfected AAT-Z in a liver-derived Huh7.5 null cell line in which AAT gene has been knocked-out (Huh7.5^null) ^82,83. Both AA-147 and AA-263 significantly increased expression of the ATF6 target chaperones GRP78/BiP and GRP94, as well as the ATF6 regulated protein disulfide isomerase PDIA4, indicating activation of the ATF6 signaling pathway/transcriptional program in these cells (Fig. S1A–B).

Figure 1. ATF6 pharmacologic activation corrects AAT-Z in liver-derived Huh7.5^null cells.

(A) Chemical structures of ATF6 activators AA-147 and AA-263. (B) Immunoblot analysis of intracellular and secreted AAT-Z in the presence or absence (vehicle control (Veh)) of AA-147 or AA-263. Huh7.5^null (AAT^−/−) cells transfected with AAT-Z plasmid were treated with AA-147 (10 μM) or AA-263 (10 μM) for 24 h. The supernatant of cell lysate and collected culture medium were analyzed by SDS-PAGE to measure the intracellular trafficking and secretion of AAT-Z. The bands of mature (M) and immature (IM) glycoforms of AAT are labeled. (C-D) Quantification of intracellular trafficking index (M/(IM+M)) (C) and the total secretion (D) of AAT-WT and AAT-Z in (B). (E) Immunoblot showing insoluble AAT-Z in lysates of cells treated with AA-147 or AA-263 as in (B). Insoluble fractions were resolubilized through sonification. (F) Quantification of the insoluble AAT from immunoblots as shown in (E). (G) The intracellular polymer in the lysates of cells treated with AA-147 or AA-263 as in (B) is measured by conformational-dependent ELISA using polymer specific antibody 2C1. (H) Immunoblot of collected culture medium from cells treated with AA-147 or AA-263 as in (B) analyzed by non-denaturing gel. (I-J) Quantification of secreted AAT polymer (I) and monomer (J) from immunoblots as shown in (H). (K) Formation of AAT-NE conjugate in response to AA-147/AA-263. Huh7.5^null (AAT^−/−) cells transfected with AAT-Z were treated by AA-147 (10 μM) or AA-263 (10 μM) for 24 h. The collected culture medium was incubated with NE for 30 min, then the AAT-NE conjugate was measured by SDS-PAGE and immunoblot analysis. (L) Quantification of AAT-NE conjugate from immunoblots as shown in (K). (M) The NE inhibitory activity of secreted AAT-Z measured by the fluorogenic substrate of NE in response to AA-147 or AA-263. (N-Q) hiPSC cells derived from homozygous ZZ patient were treated with 1, 5 or 10 μM of AA-147 for 24 h. The secreted AAT-Z monomer (N) was measured by ELISA using monomer specific antibody 16F8. The NE inhibitory activity (O) of secreted AAT-Z was measured using fluorogenic substrate of NE. The intracellular (P) and secreted (Q) polymer were measured by ELISA using polymer specific antibody 2C1. Data is presented as mean ± SD; n ≥ 3; Student’s t-test; *, p<0.05; **, p<0.01; N.S., p>0.05).

AAT has three N-linked glycosylation sites. ER to Golgi trafficking efficacy can be measured by following the processing states of these oligosaccharides. The immature, core-glycosylated isoforms (IM band) found in the ER migrates faster on SDS-PAGE than the mature, complex-glycosylated isoforms (M band) generated in the post-ER Golgi compartments preceding secretion at the cell surface (Fig. 1B) ^24,84. Treatment with AA-147 or AA-263 increases both the ER to Golgi trafficking index (M band/ (IM band + M band)) or (M/(IM+M)) (Fig. 1B and C; Fig. S1C) and the total secretion of AAT-Z (Fig. 1B and D; Fig. S1C) in Huh7.5^null cells.

The AAT-Z variant forms an intracellular polymer or aggregate that leads to gain-of-toxic pathology and liver disease ²⁸. We found that treatment with AA-147/AA-263 dramatically decreases the AAT-Z protein in the pellet of cell extracts (Fig. 1E, F), indicating that ATF6 activators decrease the insoluble AAT-Z aggregate. Consistently, using an enzyme-linked immunosorbent assay (ELISA) with a monoclonal antibody 2C1 ⁸⁵ that specifically recognizes polymer AAT, we found that treatment with AA-147 or AA-263 decreased intracellular polymer of AAT-Z (Fig. 1G). The AAT-Z polymer has been shown to be secreted from numerous cell models and can be found in serum, acting as a potent neutrophil chemoattractant, possibly contributing to COPD lung inflammation and pathology ^25,86–88. Non-denaturing gels of the collected secreted pool reveals that AA-147/AA-263 decreased the secretion of AAT-Z polymer from Huh7.5^null cells (Fig. 1H, I; Fig. S1D) and increased the level of secreted AAT-Z monomer (Fig. 1H, J; Fig. S1D). Consistently, the conformation-dependent ELISA using the polymer specific antibody 2C1 shows that the secreted AAT-Z polymer was reduced by AA-147/AA-263 treatment (Fig. S1E). In addition, the ELISA using a monoclonal antibody 16F8 that specifically recognizes monomeric AAT ⁸⁰ reveals that the secreted AAT-Z monomer was increased by AA-147/AA-263 administration (Fig. S1F).

To address whether AA-147/AA-263 impact the NE inhibitory activity of AAT, we used SDS-PAGE to capture the covalent conjugate between secreted AAT and NE before and after drug candidate treatment. Strikingly, AA-147/AA-263 administration increased the conjugates formed between NE and secreted AAT-Z around 2-fold (Fig. 1K, L), suggesting an improved activity for secreted AAT-Z as the conjugate formation inactivates NE. Using a fluorogenic substrate of NE, SucAla3-pNA (N-Succinyl-Ala-Ala-Ala-p-nitroanilide) ⁸⁹, we showed that treatment with AA-147/AA-263 increased the NE inhibitory activity of secreted AAT-Z around 1.6-fold to reach ~80% of WT activity (Fig. 1M). These results indicate that pharmacological ATF6 activators not only reduce the level of pathogenic AAT-Z polymer contributing to liver disease, but also increase NE inhibitory activity. Because doubling the NE inhibitory activity of AAT-Z could reach the range of AAT activity in the individuals carrying heterozygous M and Z alleles, who are at lower risk to develop disease ^31,90, the improvement of AAT activity by AA-147/AA-263 has potential to rescue AAT based lung disease leading to COPD. In addition to liver-derived Huh7.5^null cell line, we also observed similar correction effects of the Z-variant in the IB3 cell line, a human branchial epithelial cell with undetectable endogenous AAT background, which was either stably (Fig. S1G–I) or transiently (Fig. S1J–M) transfected with AAT-Z. AA-147 and AA-263 increased the trafficking efficiency (Fig. S1H), total secretion (Fig. S1I), monomer secretion (Fig. S1J), NE inhibitory activity (Fig. S1K), while simultaneously reducing intracellular (Fig. S1L) and secreted polymer (Fig. S1M). These results suggest that ATF6 activation by AA-147 or AA-263 targets the folding/misfolding and functional features of AAT that are conserved across different cellular secretory environments in a variety of cell lines.

In support of the relevance of the results outlined above for two heterologous cell lines transfected with AAT-Z, we also found that AA-147 treatment dose-dependently increases AAT-ZZ monomer secretion (Fig. 1N) and AAT-ZZ NE inhibitory activity (Fig. 1O) in hepatic lineages prepared from iPSCs originating from a homozygous AAT-ZZ patient (Fig. S1N). Moreover, AA-147 treatment dose-dependently reduced both intracellular (Fig. 1P) and secreted polymer levels (Fig. 1Q), demonstrating that proteostasis correction mediated by pharmacological ATF6 activation is applicable for AAT-Z produced by an endogenous promotor in a native patient-derived cellular environment.

AAT-Z correction by AA-147/AA-263 is dependent on ATF6 activation.

To understand whether the effect of AA-147/AA-263 is specific to the ATF6 pathway, we tested AA-147 (Fig.2A–B; Fig. S2A–F) and AA-263 (Fig. S2G–L) in the absence or presence of PF-429242 (10 μM) or Ceapin-A7 (10 μM). Ceapin-A7 is a specific ATF6 signaling inhibitor targeting the ATF6α branch ⁷³. PF-429242 is an inhibitor of S1P that is the protease responsible for ATF6 cleavage in the Golgi ⁹¹. We confirmed that Ceapin-A7 prevents the AA147-dependent induction of the ATF6 targets GRP78/BiP and GRP94 (Fig. S2A, B). Co-treatment with an ATF6 inhibitor such as PF-429242 or Ceapin-A7 blocks the improvement of ER to Golgi trafficking and total secretion of AAT-Z afforded by AA-147 (Fig. S2C–D) and AA-263 (S2G–H-). Consistent with these results, co-treatment with PF-429242 or Ceapin-A7 prevent the ability of AA-147/AA-263 to increase secreted AAT-Z monomer, as measured by ELISA (see Methods) (Fig. S2E; Fig. S2I). Co-treatment with PF-429242 or Ceapin-A7 also prevents the ability of AA-147/AA-263 to increase the NE inhibitory activity of secreted AAT-Z (Fig. 2A; Fig. S2J). Finally, co-treatment with PF-429242 or Ceapin-A7 blocked the effect of AA-147/AA-263 in reducing both the intracellular AAT-Z polymer accumulation (Fig. 2B; Fig. S2K) and extracellular secreted AAT-Z polymer (Fig. S2F; Fig. S2L). While AA-147 and AA-263 are structurally distinct ATF6 activators with likely different off-target activities, both AA-147 and AA-263 operate through a common mechanism by inhibiting protein disulfide isomerases mediating ATF6 ER to Golgi trafficking responsible for its activation ^59,61. As both compounds generate similar correction impact on AAT-Z that is blocked by PF-429242 and Ceapin-A7 (Fig. 1A–M; Fig. 2A–B; Fig. S2), these results demonstrate that the rescue of AAT-Z by AA-147 and AA-263 is dependent on ATF6 activation.

Figure 2. Correction of ATF6 activators on AAT-Z involves multiple mechanisms.

(A-B) NE inhibitory activity of secreted AAT-Z in conditioned media (A) and intracellular AAT-Z polymer (B) in Huh7.5^null cells transfected with AAT-Z treated with AA-147 (10 μM) and/or the ATF6 inhibitors PF-429242 (10 μM) or Ceapin-A7 (10 μM) for 24 h. (C-D) NE inhibitory activity of secreted AAT-Z in conditioned media (C) and intracellular AAT-Z polymer (D) in Huh7.5^null cells transfected with siRNA for GRP78 or GRP94, or no siRNA as native control (NC) for 24 h and then transfected with AAT-Z plasmid for 24 h for AA-147 treatment (24 h). (E-F) Intracellular polymer (E) and secreted polymer (F) of AAT-Z from Huh7.5^null cells transfected with AAT-Z treated with autophagy inhibitor Bafilomycin A1 (10 μM) in the presence or absence of AA-147 or AA-263 (10 μM) for 24 h. (G-H) Cycloheximide (CHX) chase analysis of AAT-Z intracellular polymer. Huh7.5-null cells transfected with AAT-Z plasmid were pre-treated with AA-147/AA-263 (10 μM) for 24 h, then co-treated with AA-147/AA-263 (10 μM) and cycloheximide (CHX, 100 μg/ml) in the absence (G) or presence (H) of Bafilomycin A1 (BafA1) for different times (0, 1, 2, 4 h). Cell lysate was collected to measure AAT intracellular polymer level by ELISA. (I) NE inhibitory activity of secreted AAT-Z in conditioned media from Huh7.5^null cells transfected with AAT-Z plasmid treated with autophagy inhibitor Bafilomycin A1 (10 μM) in the presence or absence of AA-147/AA-263 (10 μM) for 24 h. Data is presented as mean ± SD; n ≥ 3; Student’s t-test; *, p<0.05; **, p<0.01; N.S., p>0.05).

Given that ATF6 activators upregulate the ER chaperone proteins GRP78/BiP and GRP94 (Fig. S1A–B), we tested whether ATF6 induced correction was dependent on increased activity of these chaperones with AAT ⁹². GRP78 or GRP94 RNAi significantly reduced GRP78 and GRP94 expression, respectively, with or without AA-147 treatment (Fig. S2M). We observed that reduction of GRP78 and GRP94 expression prevented the ability of AA-147 to reduce intracellular polymerization (Fig. 2D), increase monomer secretion (Fig. S2N), or increase secreted AAT-Z NE inhibitory activity (Fig. 2C). These results indicate that ATF6 activation by AA-147 correction is dependent on ER chaperones GRP78/Bip and GRP94, which are upregulated by the ATF6 transcriptional program.

Correction of AAT-Z polymer and function by ATF6 activators involves multiple downstream mechanisms

To explore the cellular mechanisms involved in the correction on AAT-Z by ATF6 activators, we first checked for autophagy related pathways that have been shown to be responsible for the degradation of AAT-Z intracellular polymer ^93–95. To identify the role of lysosome-associated degradation, we use Bafilomycin A1 (BafA1), a lysosome inhibitor targeting the H+-ATPase enzyme that inhibits lysosome acidification and as a consequence autophagosome-lysosome fusion ^96,97. BafA1 blocks the effect of AA-147/AA-263 treatment on decreasing the levels of the intracellular and extracellular secreted AAT-Z polymer (Fig. 2E, F). Using cycloheximide (CHX) to monitor stability, AA-147/AA-263 administration was found to accelerate the degradation of intracellular AAT-Z polymer (Fig 2G) by a BafA1 sensitive pathway (Fig. 2H). In contrast, the improvement of NE inhibitory activity (Fig. 2I) and monomer secretion promotion (Fig. S2O) by AA-147/AA-263 treatment was not impacted by BafA1 co-administration. These results indicate that ATF6 activators rescue AAT-Z monomer secretion and NE inhibitory activity through potentially different downstream mechanisms from those correcting intracellular AAT aggregation, suggesting ATF6 activation separates the management of ER biogenesis of AAT from its function as a NE inhibitor. These results raise the intriguing possibility that multiple cellular pathways are enhanced by pharmacological ATF6 activation-associated transcriptional reprogramming of the secretory pathway proteostasis network, such as GRP78/Bip and GRP94 chaperone/cochaperone systems. These pathways achieve correction of AAT-Z, specifically the pathogenic load of the intracellular polymer in the liver while simultaneously restore the extracellular NE inhibitory activity in serum that is crucial to alleviate lung disease.

ATF6 activators have a broad impact on AAT variant NE inhibitory activity

Besides the prominent AAT-Z allele, there are >600 genetic variants of AAT found in the human population that contribute to diverse clinical phenotypes of AATD ^{25,28,80,98,99}. To understand the effect of ATF6 activators on genetic variation distributed across the AATD population, we collected 71 AAT variants of which ~50% have clearly defined lung and/or liver disease phenotypes ^28,80. These variants are distributed along the entire AAT polypeptide chain (Fig. 3A–C) where they could impact function-structure relationships previously defined as the clasp, breach, shutter, and gate (Fig. 3C). These function-structure relationships reflect the differential roles of the protein fold design in monomer secretion, NE inhibitory activity and intracellular polymerization.

Figure 3. Responses of AAT variants to AA-147 on NE inhibitory activity and monomer secretion.

(A) Distribution of AAT variants found in the population used in this study across the primary sequence of AAT. (B) AAT 3D structure with each secondary structural element labeled. (C) Distribution of AAT variants on the 3D structure with the alpha carbon atoms shown by spheres. The Z allele (E366K) and S allele (E288V) are labeled. AAT gate motif, breach motif, shutter motif and clasp motif are highlighted by dash circle or square. (D-E) The responses of AAT variants to AA-147 (10 μM) in Huh7.5^null cell for NE inhibitory activity (D) and secreted monomer (E). The variants were ordered by basal condition values from lowest value to highest value. WT, Z allele and S allele are labeled and highlighted by arrows. Data is presented as mean ± SD, n ≥ 3.

We first measured the NE inhibitory activity of each of these variants in the absence or presence of the ATF6 activators AA-147/AA-263 (Fig. 3D, Fig. S3A). Strikingly, treatment with either compound increases the NE inhibitory activity for most of the variants that have lower NE inhibitory activity than that observed for WT AAT treated with vehicle control (Fig. 3D, Fig. S3A) ⁸⁰. Moreover, the ATF6 activator increases monomer secretion for fewer AAT variants (Fig. 3E, Fig. S3B), implying that ATF6 activation can increase the specific activity of AAT NE inhibition by increasing the quality but not always the quantity of secreted AAT variant. NE inhibitory activity correction depends on ATF6 signal activation as the ATF6 inhibitor CeapinA7 eliminated the correction effect on NE inhibitory activity of AA-147 for multiple AAT variants (Fig. S3E, F). These results suggest the potential utility of ATF6 activators for multiple genotypes.

AAT NE inhibitory activity of variants in the basal condition shows a strong linear correlation with secreted monomer (Fig. S4B, Pearson r-value = 0.75. p = 1.70×10⁻¹⁴). Interestingly, treatment with the ATF6 activator AA-147 reduces this correlation (Fig. S4C; Pearson’s r-value = 0.51, p = 3.79×10⁻⁶) with a general improvement of the NE inhibitory activity (Fig. 3D–E; Fig. S4B vs Fig. S4C), suggesting that ATF6 may operate by uncoupling a set-point ^75,80 that manages stringency of folding in the ER to accommodate variation to yield a functional protein.

Applying GP to understand the residue-by-residue response of the AAT to ATF6 signaling

The distinctive monomer secretion (Fig. 3E) and NE inhibitory activities (Fig. 3D; Fig. S3A–B) associated with natural AAT variants distributed across the protein sequence (Fig. 3A, C) suggest the potential for a differential role of each residue in the sequence that, as a collective, contribute to the global ensemble of AAT sequence-to-function-to-structure relationships found in the AATD population. This view is currently missing from our understanding of disease states, and therefore limiting our ability to mechanistically develop more effective therapeutic approaches relevant to the clinical setting ^75–80.

In order to better understand the role of all residues shaping WT AAT fold design and how they talk to one another to generate a functional structure(s) in response to ATF6 activators, we applied GP based VSP ^75–80 (Fig. S4A). VSP is a probabilistic machine learning tool that amplifies SCV relationships linking the phenotype information of a sparse collection of variants defined by their genotype to assess function for every amino acid residue in the polypeptide sequence ^75–80. GP-based SCV builds high resolution ‘phenotype landscapes’ that can be displayed as barcodes allowing us to interpret the multi-dimensional functional relationships for all residues as an integrated collective at atomic resolution ^75–80.

Given the potential impact of sequence variation on the generation of secreted monomer and the level of NE inhibitory activity essential for normal lung function, we arrange the 71 variants in the context of their residue position in the AAT sequence (Fig. 4A, x-axis) relative to their secreted monomer levels (Fig. 4A, y-axis) and NE inhibitory activity (Fig. 4A, z-axis, color scale) in the absence (Fig. 4A, top panel) or presence of AA-147 (Fig. 4A, bottom panel). VSP first analyzes the 3-dimensional (3D) spatial relationships between every pairwise combination of variants in such a plot (Fig. 4A, black lines) to build a variogram, a metric that models as a collective the separation distances and the corresponding variance (Fig. S4D, see STAR Methods). Based on the modeling, we apply GP-based machine learning to generate the phenotype landscapes as output predicting the NE inhibitory activity (Fig. 4B, z-axis (color scale)) in the context of level of secreted monomer (Fig. 4B, y-axis) for every residue across the full-length AAT polypeptide sequence (Fig. 4B, x-axis) in the absence (Fig. 4B, left panel) or presence of AA-147 (Fig. 4B, right panel). As a probabilistic machine learning method, GP also generates an uncertainty (i.e., GP variance) for each prediction value in the landscape. We fit a two-component Gaussian mixture model over the distributions of GP generated variances across the whole landscape to separate the high confidence vs low confidence predictions (Fig. S4E, see STAR Methods). The mean of the distribution for the low GP variance is indicated in the landscapes as the outermost bold contour to reflect the regions with high confidence prediction (Fig. 4B, contours; Fig. S4E). The confidence of prediction increases along the inner contours, which are defined by the standard deviation of the distribution of the low GP variance (Fig. 4B, contours; Fig. S4E). In the basal state condition ⁸⁰, the NE phenotype landscape achieves significant prediction of the NE inhibitory activity value for any unknown point in the phenotype landscape across the entire AAT sequence based on leave-one-out cross validation (Pearson’s r = 0.62, p = 5.4 × 10⁻⁹) (see STAR Methods).

Figure 4. Mapping residue-by-residue response of NE inhibitory activity to AA-147.

(A) AAT variants are organized by their variant residue position (x-axis) normalized by the full-length polypeptide sequence, secreted monomer (y-axis) and NE inhibitory activity (z-axis, color scale) in the absence (upper panel) or presence (lower panel) of AA-147. All possible pairwise combinations of variants are illustrated by black lines. (B) Phenotype landscapes generated by GP-based VSP approach linking secreted monomer (y-axis) and NE inhibitory activity (z-axis, color scale) across the entire AAT polypeptide residue positions (x-axis) in the absence (left panel) or presence (right panel) of AA-147. The mean of the distribution of low GP variance generated by Gaussian mixture model, and the standard deviation (SD) below the mean are illustrated as contours to indicate high confidence predictions. E288V (S) and E366K (Z) are labeled in blue. (C) Residue-based NE inhibitory activity barcode derived from phenotype landscape in the absence (upper barcode) or presence (middle barcode) of AA-147. The delta (Δ) values between them are presented as the lower barcode. The assigned NE inhibitory activity values for each residue across the phenotype landscapes are averaged using the reciprocal of GP-generated variance as weight (inverse variance weighting (IVW)). The sequence regions N1, M2 and C3, previously identified to harbor multiple variants leading to defective NE inhibitory activity at basal state ⁸⁰ are indicated. The secondary structure elements of AAT sequence are indicated on the bottom. (D-H) Mapping the residue-based NE inhibitory activity in the absence (D) and presence (E) of AA-147, and their delta (Δ) values (F) to AAT 3D structures (PDB:3NE4). Highly responding structure regions are highlighted by dash circles and zoomed in (G) and (H). Variants with >40% increase of NE inhibitory activity are labeled and illustrated as sticks and balls. E228V (S) and E366K (Z) are shown in spheres and labeled.

Addition of the ATF6 activator AA-147 resulted in a dramatic change in the phenotype landscape relative to the vehicle control (DMSO) (Fig. 4B). For example, large areas with defective NE inhibitory activity (Fig. 4B, left panel, yellow-orange-red) achieve a more WT-like NE inhibitory activity (Fig. 4B, right panel, green), although there are certain regions that are less corrected, particularly at the N terminal and C-terminal region (Fig. 4B, right panel, orange-red). These results capture in an unprecedented manner the differential change in activity on a residue-by-residue basis in the AAT polypeptide sequence reflecting the change of proteostasis network capacity in the ER in response to ATF6 activation, illustrating the high plasticity of the protein fold design to management by the various pathways comprising the proteostasis network ^75–77 (Fig. 4B). The Pearson’s r-value for the leave-one-out cross validation in the presence of ATF6 activator was lower than the basal state landscape (Fig. 4B, Pearson’s r = 0.44, p = 1.5 × 10⁻⁴), a result consistent with a gain in flexibility in fold design in response to ATF6 activation captured by the general modeling of the data through the variogram ^75,76,100. For example, we observed a decrease of the ‘range’ in the variogram (Fig. S4D), a metric that describes the general length of correlated sequence in the context of secreted monomer to affect the defective NE inhibitory activity observed in the phenotype landscapes (Fig. 4B). These results are consistent with pharmacological ATF6 activation altering the folding environment of the ER, relaxing the more stringent relationships found at lower levels of the GRP78/BiP and GRP94 chaperone pathways in the native basal state to redefine the level of secreted monomer and the level of NE inhibitory activity on a residue-by-residue basis in the context of the UPR.

Assigning proteostasis function to structure

To understand how each residue impacts the fold to generate the NE inhibitory activity in response to ATF6 activator, we averaged the predicted NE inhibitory activity in the context of secreted monomer across the entire landscape for each residue using inverse variance weighting (IVW) (Fig. S4F, G, see STAR Methods) ⁸⁰. IVW prioritizes the predicted value with low uncertainty (i.e., low GP variance) over the predicted value with high uncertainty (i.e., high GP variance) by using the reciprocal of GP variance as weight during averaging (Fig. S4G, see STAR Methods) ⁸⁰. Using IVW, we obtained the most likely role of a residue contributing to the NE inhibitory activity inferred through GP based SCV relationships. These relationships reflect the activity of natural AAT variants found in the human population leading to disease (Fig. 4C, Fig. S4G). These values were plotted as a linear barcode where each residue has an assigned IVW score reflecting basal state NE inhibitory activity (Fig. 4C, upper barcode (DMSO vehicle)), its value after treatment with AA-147 (Fig. 4C, middle barcode), and the delta (Δ) value between them (Fig. 4C, lower barcode). Consistent with our previous study focusing on SCV relationships defining the basal state of the fold in the AATD population ⁸⁰, the residue-based NE inhibitory activity in the presence of DMSO vehicle reveals three major regions at the N-terminal (N1), middle (M2) and C-terminal (C3) along the primary sequence that are clustered with variants contributing to defective NE inhibitory activity (Fig. 4C, upper barcode). Strikingly, AA-147 treatment generally increased the NE inhibitory activity for most variants in each of the 3 clusters critical for native basal state fold design ⁸⁰ (Fig. 4B; Fig. 4C, middle and lower barcodes). In particular, the C-terminal half of N1 region (hD-hF), most of the M2 region (s4C-hH) and the N-terminal half of the C3 region (s5A-RCL-s1C) showed a dramatic impact on correction leading to near WT-like NE inhibitory activity after treatment with AA-147 (Fig. 4B; Fig. 4C).

To understand the structural mechanisms directing the activity response to AA-147 at atomic resolution, we constructed a ‘functional structure’ ^75,80 by mapping the assigned NE inhibitory activity for each residue in the absence (Fig. 4C, upper barcode) or presence of AA-147 (Fig. 4C, middle barcode), as well as the delta (Δ) NE inhibitory activity in response to AA-147 treatment (Fig. 4C, lower barcode) to AAT structure (PDB: 3NE4) (Fig. 4D, E, F). Comparing the functional structure in the presence of AA-147 (Fig. 4E) to the DMSO control vehicle state (Fig. 4D) reveals that the activation of ATF6 signaling pathway differentially impacts the overall AAT structure. Specifically, the delta (Δ) structure (Fig. 4F, G, H) shows several highly responsive structural regions to AA-147 that includes the M2 sequence region (Fig. 4C) located at the ‘top’ of AAT molecule near the RCL and AAT-Z variant in the 3D structure (Fig. 4F, circle 1). It comprises s3C and s4C that have been previously indicated as ‘gate’ area regulating the loop-sheet formation mechanism that is important for AAT function following secretion from the liver (Fig. 4G) ^101,102. It also harbors s1B-s3B in β-sheet B and hG-hH where another frequent mild AATD variant AAT-S (E288V) is located (Fig. 4G). These structural regions form putative latch interactions that are suggested to link the mobility of the gate area to the release of the RCL and s1C leading to the loop-sheet insertion responsible for NE inhibition ^101–103. Furthermore, the β-sheet A (s1A-s6A) in which RCL inserts as the fourth β-strand (s4A) during the loop-sheet transition is also highly responsive to AA-147 administration (Fig. 4F, circle 2; Fig. 4H). These results suggest that activation of ATF6 pathway through AA-147 treatment likely adjusts the post-secretion folding and stability of the loop-sheet insertion mechanism sensitive to variation in inherited sequences triggering disease through yet unknown cellular mechanisms to improve the NE inhibitory activity of the secreted AAT.

In addition to the gate and β-sheet A regions, the C-terminal half of N1 region (hD-hF) also exhibits a strong response to AA-147 treatment (Fig. 4C, 4H). We have previously shown that most AATD variants in the C-terminal half of N1 region (hD-hF) do not contribute to polymer formation in the cell, but rather cause rapid degradation of monomer ⁸⁰. These results now suggest AA-147-mediated ATF6 activation may improve the monomer load in the ER for export as an additional feature of ATF6 activation that contributes to the increase the NE inhibitory activity. Interestingly, the weaker response regions - the N-terminal half of N1 (hA-hC) and the C-terminal half of C3 (s4B-s5B) (Fig. 4C) interact with each other through the interactions between s6B from N1 and s5B from C3 (Fig. 4F; Fig. 4G). This result reveals that the long range folding events between the N-terminal and C-terminal structural elements are more challenging to correct by AA-147. Overall, these results reveal that ATF6 activation and the corresponding changes in ER associated proteostasis components and their levels may use different mechanisms to reshape the sequence-to-function-to-structure relationships responsible for correction of secreted AAT on a residue-by-residue basis.

ATF6 activation selectively improves monomer secretion

While the above SCV relationships focused on a map defining the role of NE inhibitory activity for every residue in AAT, to understand the impact of ATF6 activation over the entire AAT polypeptide sequence in terms of monomer secretion, we used the NE inhibitory activity as the y-axis feature to predict the role of each residue in monomer secretion as the z-axis feature of phenotype landscape before and after treatment with AA-147 (Fig. 5; Fig. S5). AA-147 dramatically reduces the correlation range in the variogram (Fig. S5B) and the area defining the high confidence regions in the landscape (Fig. S5C, D). Moreover, AA-147 reduces the prediction accuracy (Pearson’s r = 0.6, p = 2.4 × 10⁻⁸ in the absence vs Pearson’s r = 0.3, p = 0.01 in the presence of AA-147). These results suggest that activation of ATF6 by AA-147 administration largely uncouples the SCV relationships linking NE inhibitory activity to monomer secretion across the polypeptide sequence, reflecting the change in plasticity of the fold in response to the altered proteostasis enhanced ER folding environment.

Figure 5. Mapping residue-by-residue response of monomer secretion to AA-147.

(A) Residue-based monomer secretion barcodes derived from the phenotype landscapes through IVW in the absence (upper barcode) or presence (middle barcode) of AA-147. The delta (Δ) values are presented as the lower barcode. (B-F) Mapping the residue-based monomer secretion in the absence (B) and presence (C) of AA-147, and their delta (Δ) values (D) to AAT 3D structure. (E-F) The regions that were highlighted in Fig. 4G–H are zoomed in for comparison between the responses to NE inhibitory activity (Fig. 4G–H) and monomer secretion (E-F).

To provide barcode view of these results we averaged the level of secreted monomer for each residue using IVW to build the residue-based monomer secretion barcode in response to AA-147 (Fig. 5A) and mapped these values on AAT 3D structures (Fig. 5B–D). The results show that improved monomer secretion in response to AA-147 is highly selective with many residues showing no or very little correction (Fig. 5). This is in striking contrast to the NE inhibitory activity where most of the residues highly respond to AA-147 (Fig. 4). Notably, these results suggests that the increase of NE inhibitory activity is largely not due to a corresponding increase in secreted monomer.

The delta (Δ) value for monomer secretion highlights the specific regions where AA-147 dosing improves monomer secretion. For example, residues in the signal peptide found at the N-terminus of the AAT sequence are responsive to AA-147 treatment (Fig. 5A, lower barcode), suggesting the possibility that ATF6 activation can impact either translation and/or translocation steps of the nascent polypeptide into the ER to improve the secretion of functional monomer. The C-terminal half of N1 region (s2A-hE) also shows a high correction of monomer secretion (Fig. 5A, lower barcode; Fig. 5D, circle 2; Fig. 5F). As we have previously shown that this region in the native basal state does not contribute to polymer formation, but triggers rapid degradation of monomer ⁸⁰, these results suggest that ATF6 improves monomer secretion by prevention of delivery of AAT to ERAD. Additional regions that show a high response to AA-147 leading to monomer secretion include s1B-s2B-s3B in M2 region and s5B at the end of C3 region (Fig. 5A, lower barcode). They are clustered together through β-sheet B (Fig. 5D, circle 1; Fig. 5E), suggesting AA-147 can impact the folding of β-sheet B to improve monomer secretion. Surprisingly, the gate area, defined by s3C and s4C, that is highly corrected in terms of NE inhibitory activity (Fig. 4G), does not show correction of monomer secretion (Fig. 5E). Consistent with this observation, the highly selective correction of monomer secretion (Fig. 5) compared to the broad correction of NE inhibitory activity (Fig. 4) supports the interpretation that AA-147 improves activity in the secreted monomer mainly through an increase in the capacity to perform the activity of the rescued AAT variant rather than a direct increase in the AAT monomer level.

Global response of intracellular polymerization to ATF6 activators

AAT-Z misfolding leads to pathogenic intracellular polymer ER accumulation in liver hepatocytes ^25,28,46. Moreover, recent clinical observations suggest serum polymer levels can be used as a surrogate biomarker for the extent of liver disease ^104,105. To address how the spectrum of AAT variants found in the population impact intracellular polymerization and how ATF6 activation might alleviate this aggregated state, we measured the level of intracellular polymer of 71 variants in the absence or presence of ATF6 activators (Fig. 6A for AA-147; Fig. S3C for AA-263). Strikingly, both activators decrease the intracellular polymer for most of the variants compared to the basal state (Fig. 6A, Fig. S3C), raising the possibility that ATF6 activators can rescue AATD disease phenotypes triggering liver disease. Consistent with this conjecture, ATF6 signaling inhibitor Ceapin-A7 significantly eliminated the ability of AA-147 to reduce intracellular polymer accumulation as observed for E366K (Z) (Fig. 2B), and also for F76del, G249R, S allele (E288V) and H358D variants which present with high polymer accumulation in the ER in the basal condition (Fig. S3G) ⁸⁰. Importantly, in cell-based measurements (Fig. 6A), AAT intracellular polymer pools generated for the variants found in the AATD population show a strikingly strong correlation with extracellular polymer in the absence (Fig. S6A; Fig. S6B, left panel, Pearson’s r = 0.8, p = 6.20×10⁻¹⁷) or presence (Fig. S6B, right panel, Pearson r = 0.77, p = 6.27×10⁻¹⁵) of AA-147. These results raise the possibility that ATF6 activation can reduce the level of secreted polymer through modulation of the intracellular polymer, and validate the use of secreted polymer as a potentially universal biomarker of variant AAT disease ¹⁰⁴.

Figure 6. Mapping residue-by-residue response of intracellular polymer to AA-147.

(A) The responses of AAT variants to AA-147 (10 μM) in Huh7.5^null cell for intracellular polymer. (B) Residue-based intracellular polymer barcodes derived from the phenotype landscapes through IVW in the absence (upper barcode) or presence (middle barcode) of AA-147. The delta (Δ) values between them are presented as the lower barcode. (C-F) Mapping the residue-based intracellular polymer in the absence (C) and presence (D) of AA-147, and their delta (Δ) values (E) to AAT monomer structure. The highly responding region to AA-147 for intracellular polymer is zoomed in (F). Variants with >20% decrease of intracellular polymer are labeled and illustrated as sticks and balls. (G-I) Mapping the residue-based intracellular polymer in the absence (G) and presence (H) of AA-147, and their delta (Δ) values (I) to AAT polymer structure (PDB: 3T1P).

To understand the mechanism of polymer correction by ATF6 activation, we applied VSP to profile the relationships between secreted polymer and intracellular polymer for each residue across AAT polypeptide sequence before and after treatment of AA-147 (Fig. S6C–F). Both the range and plateau value in the variogram modeling are decreased by AA-147 (Fig. S6D), consistent with a general decrease of both intracellular and secreted polymer (Fig. S12D). The range in both cases span the entire sequence (Fig. S6D, range >1), suggesting that although AA-147 reduces the impact of variant diversity in polymer formation, the level of secreted polymer remains strongly correlative with level of intracellular polymer across the full-length sequence of AAT. The intracellular polymer phenotype landscapes in the absence and presence of AA-147 both achieve significant prediction in the leave-one-out cross validation results (Pearson’s r = 0.67, p = 3×10⁻¹⁰ (DMSO); Pearson’s r = 0.6, p = 2×10⁻⁸ (AA-147)). Consistent with these results, VSP accurately predicted the phenotype responses for a variant K283I that was not used as input (Fig. S7; see STAR Methods). These results indicate that the phenotype landscapes are able to generate high confidence predictions to assess the residue-by-residue response mechanisms to ATF6 activators across the entire AAT sequence.

We generated residue-based intracellular polymer barcodes in the absence (Fig. 6B, upper barcode) or presence of AA-147 (Fig. 6B, middle barcode), as well as the delta (Δ) value between them (Fig. 6B, lower panel) using IVW. These values were mapped to AAT monomer structure (Fig. 6C–F) and polymer structure (Fig. 6G–I) ^46,106 to illustrate the responsive structural regions. Consistent with our previous study ⁸⁰, the intracellular polymer barcode in the presence absence of AA-147 is highly consistent with the recent C-terminal model of native AAT-Z variant polymer isolated from patient hepatocytes expressing the Z-variant aggregate ⁴⁶, where not only the s5A-RCL in region C3 forms a loop-sheet insertion, but s4B-s5B in the C3 region inserts into the β-sheet B in another molecule formed by s6B in N1 and s1B-s2B-s3B in M2 (Fig. 6G).

Strikingly, AA-147 treatment decreases the intracellular polymer for most of the residues in these regions (Fig. 6B, 6H, 6I). Specifically, we found interacting residues between s6B (e.g., F76 and S77) and s5A (e.g., H358 and A360) that show very high responses to AA-147 (Fig. 6F). These residue interactions have been shown to be critical for the opening of β-sheet A for loop-sheet insertion ^25,39, suggesting AA-147 adjusts the loop-sheet mechanism to reduce the polymer load. As a result, AAT-Z and RCL (s4A) also show high responses to AA-147 leading to reduced polymer (Fig. 6I). In addition, s5B found at the C-terminal that interacts with s6B at the N-terminal is a high response region (Fig. 6F, 6I), suggesting that ATF6 activation modulates the misfolding of C-terminal region in response to variants, possibly through late co-translational events, and/or post-translational proteostasis mediated events to reduce the polymer formation. The different mechanisms highlighted by GP analysis to rescue the intracellular polymer load suggest that the restoration of proteostasis through ATF6 activation is a residue specific event that differentially re-shapes residue-residue connections that contribute to polymerization in response to variant challenge.

A global map of residue contributions driving AAT form versus function

To address globally the relationships between intracellular polymer formation, monomer secretion, and NE inhibitory activity at residue-by-residue resolution in response to ATF6 activators, we overlaid the residue-based Δ value for each of these features in response to AA-147 administration (Fig. 7A; gray, magenta, cyan dots, respectively). Interestingly, the correction of monomer secretion, polymerization and NE inhibitory are very similar at the extreme N-terminal region of AAT sequence (Fig. 7A, residues 1–100), suggesting that for variants that impact the initial co-translational step, the improvement of NE inhibitory activity is largely due to an increase of secreted AAT monomer level. After residue ~100, the correction of NE inhibitory activity is much higher than the correction of monomer (Fig. 7A, compare magenta dots with gray dots indicating residue). Moreover, there are regions in the AAT sequence where ATF6 activation has no impact on polymer levels given that variants in these regions do not contribute to intracellular polymerization in the native basal state (Fig. 6B) ⁸⁰. Surprisingly, these regions show high NE inhibitory activity correction (Fig. 7A, compare magenta dots with cyan dots). In contrast, at the C-terminus of the AAT sequence (residue >~400), we capture a distinct role for ATF6 activators in the differential management of AAT physical and functional states. Here, the prediction of NE inhibitory activity correction is similar or lower than the correction of intracellular polymer and monomer secretion (Fig. 7A, compare magenta with cyan circles). These results suggest that there is additional functional role(s) of the C-terminus of AAT that contributes to NE inhibitory activity that cannot be corrected by AA-147.

Figure 7. Residue-by-residue differential response of NE inhibitory activity, monomer secretion and intracellular polymer to AA-147.

(A) Overlay of the delta (Δ) value between the DMSO vehicle and AA-147 states for the residue-based NE inhibitory activity (magenta), monomer secretion (gray) and intracellular polymer level (cyan). Positive delta (Δ) values represent improved NE inhibitory activity, increased monomer secretion and decreased intracellular polymer. N1, M2 and C3 sequence regions in rich of variants leading to defective NE inhibitory activity at basal state are indicated. (B) Residue-by-residue NE inhibitory activity to monomer ratio in response to AA-147. The NE inhibitory activity and secreted monomer ratio is computed for both DMSO and AA-147 states for each residue. The delta (Δ) value of the activity to monomer ratio in response to AA-147 for each residue is plotted. (C) Mapping the delta (Δ) of activity to monomer ratio to AAT monomer structure. The gate area is zoomed. (D) Mapping the delta (Δ) of activity to monomer ratio to the complex structure of AAT-elastase. (E) A schematic figure illustrating the impact of pharmacological ATF6 activation on the fold and function of AAT. GP-based profiling reveals that pharmacological ATF6 activation targets different sequence regions to differentially rescue the phenotypes of AAT variants, for example, targeting the structural elements responsible for the C-terminal polymerization model to reduce polymer, assisting the assembly of β-sheet B to improve monomer secretion, and managing the gate area to increase the NE inhibitory specific activity. We posit that the remodeling of ER proteostasis network components such as the GRP78/GRP94 chaperone systems and the ER-specific autophagy pathways by ATF6 activators generates a new function-structure state for AAT variant, referred to as Z* state, that can coordinate the correction of both the gain-of-toxic liver disease and loss-of-function lung disease. The SCV-based convertible state of the fold managed by proteostasis provides a common framework on which biology builds function-structure relationships in response to natural selection.

To globally illustrate the differential responses between NE inhibitory activity and secreted monomer, we plot the response to AA-147 treatment based on the delta (Δ) of the ratio between NE inhibitory activity to AAT monomer levels to generate a metric that reports on ‘specific activity’ of the secreted AAT monomer (Fig. 7B, orange). We find the changes in the specific activity of N-terminal sequence region in response to AA-147 are mild (Δ<25%). The major increase in specific activity in response to AA-147 (Δ>25%) occurs following residue ~100 (Fig. 7B) to residue ~400. The region with the highest increase in the activity to monomer ratio in response to AA-147 is s4C in the gate area that regulates the loop-sheet insertion (Fig. 7A, 7B; red box) ^101,102, suggesting an unanticipated role of proteostasis system in managing this solvent exposed structural area (Fig. 7C, circle 1) to improve the specific activity of the protein fold design. Indeed, many structural regions associated with the loop-sheet insertion mechanism show high increase of activity to monomer ratio in response to AA-147 (Fig. 7B–C). These include: 1) s5A-RCL-s1C at the N-terminal half of C3 that forms the loop-sheet insertion containing AAT-Z; 2) β-sheet A where the RCL inserted as s4A; 3) s3B-hG-hH in M2 that forms latching interactions with the gate area ^101–103 containing AAT-S; and 4) hD-s2A-hE-s1A-hF at the C-terminal half of N1 that has been shown to form a cavity that regulates loop-sheet insertion (Fig. 7D, circle 2) ^107–109. The latter structural region is close to the NE substrate in the complex which shows significant structural changes in response to NE capture (Fig. 7D; circle 2 vs Fig. 7C; circle 2), suggesting that these structural dynamics are managed by ATF6 activators to improve secreted AAT function.

These results suggest that activation of the ATF6 proteostasis network largely prioritizes folding of AAT variants to improve function over those features of fold design that contribute to polymerization and/or ER export, suggesting the fold is managed by proteostasis from a SCV based quality ‘system’ ⁷⁸ perspective leading improvement of activity for >99% of the AATD population affected by differential progression of lung and liver disease.

Discussion

The residue-residue interactions contributing to the folding, misfolding and function of a polypeptide chain in the cell is highly dynamic. By using genetic variants in the population through GP-based VSP analysis ^75–80, we found that the SCV relationships linking every residue in AAT polypeptide chain as a functional collective is dynamically remodeled by pharmacological activation of the UPR-associated transcription factor ATF6. ATF6 activators rewire these SCV connections in a residue specific fashion through multiple proteostasis pathways on a residue-by-residue basis to generate a striking differential impact on the rescue of AAT misfolding and function leading to decreased intracellular polymerization, increased secreted monomer, and unprecedented, an improved NE inhibitory specific activity. Therefore, pharmacological activation of ATF6 pathway presents a unique strategy to simultaneously correct the liver aggregation-associated gain-of-toxic function and the loss-of-function in the lung disease phenotypes for AATD population, providing an unanticipated view for the role of proteostasis in the covariant management of human biology.

Our previous work applying VSP to understand basal functional state of AAT ⁸⁰ revealed that it is principally organized by 3 clusters, N1, M2, and C3 that span the polypeptide sequence. Each cluster differentially contributes to the coupled monomer secretion and NE inhibitory activity required for lung function with a more restricted impact on intracellular polymerization responsible for liver disease ⁸⁰. These 3 clusters connect with each other through β-sheet B to determine the intracellular polymerization ⁸⁰, consistent with the C-terminal polymerization model revealed by the cryo-EM study of in vivo AAT polymer ⁴⁶. Here, we found that the β-sheet B, especially the interactions between s6B from N1 and s5B from C3, are highly responsive to ATF6 activators in terms of reducing the polymer level (Fig. 7E). In addition, s5A-RCL region critical for the loop-sheet mechanism is also highly responsive. By applying GP analysis through VSP, we reveal that the structural elements required for different steps in the C-terminal polymerization model are manageable by ATF6 signaling. These steps include the dissociation of interactions between s5B and s6B, the loop-sheet insertion of RCL as s4A and insertion of s4B and s5B into β-sheet B of another molecule (Fig. 7E). Moreover, our analysis based on manipulation of the native ATF6 signaling pathway in the context of natural variants triggering disease, revealed the role for the ER specific autophagy-lysosome pathway under UPR activation for polymer management. These results are distinct from the effects of overexpression of the active ATF6(1–373), where autophagy was not induced leading to degradation of AAT-Z by ERAD ¹¹⁰, potentially owing to the non-physiologic levels of UPR activation observed under these conditions ¹¹¹. The impact of AA-147/AA-263 is also different from the effect of the general and toxic UPR activators tunicamycin and thapsigargin that were shown to increase the immobilization of AAT-Z, both results likely reflecting unnatural disruption of the ER folding environment ¹¹². In a recent study using a Huh7.5Z cell line wherein the endogenous AAT-WT gene was edited to AAT-Z through CRISPR/Cas9 or hepatocytes from transgenic mouse model expressing human AAT-Z gene, sustained ATF6 activation was observed ¹¹³. This result suggests that pharmacological ATF6 activators may further enhance the ATF6-adapted capacity of the ER proteostasis environment being used by hepatocytes to improve AAT-Z folding and function ^{2,59,61–64,66–68}.

GP based residue-by-residue profiling of the response to ATF6 activators reveals that the improvement of monomer secretion and activity for AATD does not necessarily require changes in the polymer level, suggesting that ATF6 activators adjust proteostasis pathways that can uncouple the correction of polymer accumulation from improvement of secreted monomer and NE inhibitory activity. This is consistent with the result that an autophagy inhibitor only blocks the intracellular polymer correction, but does not impact the improvement of monomer and NE inhibitory activity in response to ATF6 activators (Fig. 7E). Strikingly, we found that ATF6 activation generates a significantly higher response of NE inhibitory activity over the correction of monomer secretion for the majority of AAT sequence regions. Specifically, the gate area comprising s4C shows the biggest change in the NE inhibitory activity to monomer ratio in response to ATF6 activators revealed by VSP analysis. The improvement of AAT-Z NE inhibitory activity by ATF6 activators requires both GRP78 (ER paralog of cytosolic Hsp70) and GRP94 (ER paralog of cytosolic Hsp90) chaperone-cochaperone systems (Fig. 7E). These results suggest that the open and close status of the gate area regulating the loop-sheet mechanism required for AAT function ^101–103 can be modulated through proteostasis pathways sensitive to ATF6 activators in a fashion in which the improved activity is retained post-secretion. Understanding how different proteostasis pathways manage folding and function features contributed by each residue in the AAT sequence can help guide precision management of variant load in an individual.

Interestingly, the convertible feature of the AAT protein fold revealed by GP-based analysis in response to the remodeling of proteostasis environment by ATF6 signaling is reminiscent of the protein fold conversion seen in prion disease. The cellular monomer isoform of PrP^C is secreted through the same pathway as AAT but, unlike AAT, is GPI-anchored on the extracellular surface. In rare cases, it can be passaged and converts to the toxic, oligomeric species PrP^Sc triggering human neurodegenerative disease ^114–118. Similarly, the wild-type AAT monomer, referred to as the M strain, can adopt an alternative and reversable polymerization-prone conformation (M*) that can co-polymerize with AAT-Z variant in the patient ^42,119,120. Based on our results, we suggest that pharmacological activation of ATF6 generates a new proteostasis state with improved specific activity which we refer as the Z* state (Fig. 7E). This state provides flexibility to bias the secreted AAT variant to the functional state. Furthermore, both the intracellular and secreted polymer is reduced in Z* state by ATF6 activation, consistent with observations that adjustment of proteostasis in the ER can impact both intracellular and extracellular aggregation ^{12,59,65–67,121}. The ability of prevent aggregation through proteostasis in protein misfolding/amyloid disease is consistent with the growing interest of managing systemic ^{11,12,62,65–67,121,122} and neurodegenerative aging-related diseases ^{6,115,117,118,123–125} using chaperone technologies. These results lead us to posit that biology operates on a common covariant framework ⁷⁵ where proteostasis ^{2,4,6,8,10–12,16,69,124–126} serves as a normal integrated facet of the protein fold to maintain its highly convertible function-structure relationships in both normal physiology and pathology as part of a more comprehensive ‘quality system’ ⁷⁸ that manages the diversity in the population.

By integrating the natural variant response triggering AATD through GP-based VSP analysis, we illustrate the covariant design of the protein fold and of how each residue is managed uniquely by proteostasis in the context of the surrounding residues to differentially impact the folding/misfolding, secretion and function along different cellular compartments in the secretory pathway. Understanding the covariant design of these interactions based on GP is key to develop optimized therapeutics not only for AATD but for many other complex diseases including cystic fibrosis and Niemann-Pick C1 ^75–81, where the problem protein harbors multiple variants that can initiate both gain-of-function and loss-of-function mechanisms. Past efforts to develop therapeutics for AATD have lacked an understanding of the importance of pliable state of protein directing folding, misfolding and function on a residue-based resolution, resulting in limited progress ^{24,34,41,51–54,84,93,95,108,127,128}. Learning how ATF6, and the related IRE1 and PERK UPR signaling pathways can tune proteostasis to manage sequence-to-function-to-structure relationships through GP-based analysis will help guide not only the precision management of the patient ^75–81, but provide critical insights into the likely universal role of proteostasis in the covariant management of the dynamics of the protein fold during natural selection in response to the environment ^{2,4,8–11,69,75,129}.

Limitations of the study

The impacts of pharmacological ATF6 activators AA-147 and AA-263 on AAT variants were characterized in cell-based models in this study. Animal models of AATD^74,130 can be used to confirm the correction effect of the ATF6 activators to set stage for future clinical trials.

STAR Methods:

RESOUCE AVILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, William E. Balch (webalch@scripps.edu)

Materials availability

All materials generated in this study are available from the lead contact with a completed Material Transfer Agreement.

Data and code availability

• All the input source data and output files have been deposited at Mendeley database and are publicly available as of the date of publication. DOIs are listed in the key resources table.

• All original code has been deposited at Mendeley database and is publicly available as of the date of publication. DOIs are listed in the key resources table.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals and Antibodies
AA-147 and AA-263	Plate et al. 59	Provided by Dr. Jeffery W. Kelly lab and Dr. R. Luke Wiseman lab
Ceapin-A7	Sigma-Aldrich	Cat # SML2330
PF-429242	Sigma-Aldrich	Cat # SML0667
Bafilomycin A1	Abcam	Cat # 246689
Cycloheximide	Sigma-Aldrich	Cat # 01810
Complete-Mini protease inhibitor	Roche	Cat # 11836170001
goat anti-human AAT polyclonal antibody (80A)	ICL. Inc	Cat # GCYT-80A
mouse anti-human AAT monoclonal antibody 16F8	This paper	Scripps Research Antibody Development and Production Core
mouse anti-human AAT monoclonal antibody 2C1	Hycult Biotech	Cat # HM2289
GRP78 antibody	Abcam	Cat # ab108615, RRID:AB_10890641
GRP94 antibody	Abcam	Cat # ab238126
PDIA4 antibody	Abcam	Cat # ab155800
Goat anti-Mouse HRP antibody	Thermo Fisher Scientific	Cat # 32230, RRID:AB_1965958
Goat anti-Rabbit HRP antibody	Thermo Fisher Scientific	Cat# 32260, RRID:AB_1965959
Mouse anti-Goat HRP antibody	Thermo Fisher Scientific	Cat# 31400, RRID:AB_228370
Human neutrophil elastase	Innovative Research	Cat # IHUELASD100UG
Neutrophil elastase fluorescence substrate 2Rh110	Cayman Chemical	Cat # 11675
polystyrene high binding plate	Corning	Cat # 3690
FuGENE6 transfection reagent	Promega	Cat # E2691
Lipofectamine RNAiMAX	Thermo Fisher Scientific	Cat # 13778150
GRP78/Bip siRNA	Thermo Fisher Scientific	siRNA ID. 145249
GRP94 siRNA	Thermo Fisher Scientific	siRNA ID. 119656
DMEM medium	Corning	Cat # 15-013-CM
F12 medium	Sigma-Aldrich	Cat # N6658-500ML
LHC-8 medium	Thermo Fisher Scientific	Cat # 12679-015
Fetal bovine serum	Thermo Fisher Scientific	Cat # SH30396.03
penicillin streptomycin (P/S)	Thermo Fisher Scientific	Cat # 15140-122
L-glutamine	Thermo Fisher Scientific	Cat # 25030-081
Non-denaturing PAGE gel	Bio-Rad	Cat # 4561094DC
Native sample buffer	Bio-Rad	Cat # 1610738
Tris/Glycine buffer	Bio-Rad	Cat # 1610734
1-step Ultra TMB-ELISA	Thermo Fisher Scientific	Cat # 34029
Commercial assays kits
DNA purification kit	QIAGEN Inc	Cat # 27106
Experimental models: Cell lines
AAT knock out liver cell Huh7.5null (AAT−/−)	Khodayari et al. 82, 83	Provided by Dr. Mark L. Brantly, University of Florida
PiZZ3 hiPSC	Kaserman et al. 131, 132	Provided by Dr. Andrew A. Wilson, Boston University
IB3-Z cell	Bouchecareilh et al. 24	Provided by T. Flotte, University of Massachusetts Medical School.
Bacterial strains
DH5α competent cells	Intact Genomics	Cat # 1013
Recombinant DNA
AAT variants DNA	Quintara Bioscience	N/A
Deposited Data and Code
NE inhibitory activity measurements of AAT variants	This paper	Mendeley data: http://doi.org/10.17632/4xyt372py7.1
Secreted monomer measurements of AAT variants	This paper	Mendeley data: http://doi.org/10.17632/4xyt372py7.1
Intracellular polymer measurements of AAT variants	This paper	Mendeley data: http://doi.org/10.17632/4xyt372py7.1
Secreted polymer measurements of AAT variants	This paper	Mendeley data: http://doi.org/10.17632/4xyt372py7.1
Input source data, R-code scripts and output files to generate the phenotype landscapes, phenotype barcodes and functional structures.	This paper	Mendeley data: http://doi.org/10.17632/4xyt372py7.1
Software and Algorithms
Gstat	R-package	https://cran.r-project.org/web/packages/gstat/index.html
Mclust	R-package	https://cran.r-project.org/web/packages/gstat/index.html
Originpro 2020b	Originlab	https://www.originlab.com/
Pymol	Schrodinger, LLC	https://pymol.org/2/
ImageJ	National Institute of Health	https://imagej.nih.gov/ij/index.html

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines and cell culture.

Endogenous AAT knock out liver cell Huh7.5^null (AAT−/−) (origin: human, male, 57-year-old) is a gift from Dr. Mark Brantly (University of Florida College of Medicine, Gainesville, FL). Huh7.5^null (AAT−/−) cell was cultured in DMEM/F12 medium with 10% (v/v) FBS, 100 μg/ml P/S and 2 mM L-glutamine. IB3 cells (origin: human, male, 7-year-old ) and IB3 cells stability transfected with AAT-Z were provided by T. Flotte, University of Massachusetts Medical School and were cultured in LHC-8 with 10% (v/v) FBS, 100 μg/ml P/S. iPSCs from AATD patients carrying the ZZ variant was obtained from the Dr. Andrew Wilson, Center for Regenerative Medicine of Boston University and Boston Medical Center, Boston MA. The iPSCs were differentiated into hepatocytes following published protocols by the Wilson lab ¹³¹. PiZZ3 iPSC line ¹³² was selected and used in this study. PiZZ3 was generated from peripheral blood mononuclear cells (PBMC) of a 65 year old male who experienced moderate centrilobular emphysema with mild liver disease. Briefly, iPSC cells were plated on Matrigel-coasted plates for differentiation. After 4 days of endoderm differentiation, the cells were split, and lineage was validated with flow cytometry. The cells were further differentiated for 24 days, when the cells were treated with UPR activators. After a 24 h compound treatment, the culture media was replaced and conditioned for 3 h prior to assays to assess the monomer and polymer secretion, and NE inhibitory activity. The cell lysate was collected to measure the level of intracellular polymer.

METHOD DETAILS

Reagents and antibodies.

ATF6 activators AA-147 and AA-263 were provided by the labs of Jeffery W. Kelly and R. Luke Wiseman ⁵⁹. ATF6 inhibitor Ceapin-A7 was purchased form Sigma-Aldrich (No. SML2330, St. Louis, MO). S1P inhibitor PF-429242 was purchase from Sigma-Aldrich (No. SML0667, St. Louis, MO). Autophagy inhibitor Bafilomycin A1 was purchased from Abcam (No. 246689). Complete-Mini protease inhibitor Cocktail used for cell lysate was purchased from Roche (No. 11836170001). The goat anti-human AAT polyclonal antibody (80A) used in ELISA assay plate coating and for immunoblotting was purchased from ICL. Inc (No. GCYT-80A). The mouse anti-human AAT monoclonal monomer recognize protein antibody (16F8) was generated by Scripps Research Antibody Development and Production Core. The mouse anti-human AAT monoclonal polymer protein recognize antibody (2C1) was purchased from Hycult Biotech (No.HM2289, Wayne, PA). GAPDH antibody was purchased from abcam (No. ab8245). GRP78 antibody and GRP94 antibody and PDIA antibody were purchased from abcam (No. ab108615, ab238126 and ab155800). HRP secondary antibodies were purchased from Thermo Fisher Scientific. Human neutrophil elastase (NE) was purchased from Innovative Research (IHUELASD100UG, Novi, MI). NE fluorescence substrate 2Rh110 (Z-Ala-Ala-Ala-Ala) was purchased from Cayman Chemical (item No. 11675, Ann Arbor, MI). Clear flat bottom polystyrene high binding 96-well plate was purchased from Corning (Ref 3690). DNA purification kit was purchased from QIAGEN Inc (Valencia, CA). Cell transfection reagent FuGENE6 was purchased from Promega (Madison, WI). Cell transfection reagent Lipofectamine RNAiMAX was purchased from Thermo Fisher Scientific. GRP78 and GRP94 siRNA were purchased from Thermo Fisher Scientific. DMEM cell culture medium was purchased from Corning; F-12 culture medium was purchased from Sigma-Aldrich; and LHC-8 cell culture medium was purchased from Thermo Fisher Scientific. FBS, penicillin streptomycin (P/S) and L-glutamine were purchased from Thermo Fisher Scientific. Non-denaturing gel, Tris/Glycine buffer and native sample buffer were purchased from Bio-Rad (No. 4561094DC, No. 1610734 and No. 1610738).

AAT variants vectors and cell transfection.

AAT variants were collected from different databases as described in ⁸⁰. AAT variants DNA constructs in pcDNA3.1 (+) vector were generated by Quintara Bioscience (Cambridge, MA). The plasmid transfection was performed using FuGENE6 following the manufacturer protocol. Cells were treated with compounds 24 h after transfection. The siRNA transfection was performed using Lipofectamine RNAiMAX following the manufacturer’s protocol.

SDS-PAGE and non-denaturing gel immunoblots.

For AAT variants expression, AAT variants plasmids were transiently transfected in cells (Huh7.5^null cell) for 24 h. For culture medium harvest, cells were washed with 1xPBS and changed to fresh serum-free medium for 3 h. For cell lysate preparation, after transfection and treatment, whole-cell lysate was harvested with cell lysis buffer (50 mM Tris-HCL, 150 mM NaCl, 1% (v/v) Triton X-100 with complete-Mini protease inhibitor cocktail). Collection of cell supernatants was performed by 30 min centrifugation at 4°C. The total protein concentration of cell lysate was quantified by Bradford assay. For the SDS-PAGE immunoblot assay, culture medium and cell lysate samples were resuspended in SDS sample buffer and incubated at 95 °C for 5 min. For non-denaturing SDS-PAGE immunoblot assay, culture medium and cell lysate were prepared by native sample buffer. Equal protein amounts were loaded for each sample for immunoblot assay. A nitrocellulose membrane was used for sample transfer.

AAT specific conformation (monomer or polymer) ELISA assay.

For AAT specific conformation (monomer or polymer) ELISA assays, Huh7.5^null cells were planted in 96-well plates at 2×10⁴ cells per well. AAT plasmid variants were transfected at 0.2 μg/well following the FuGENE6 protocol. After 24 h transfection, cells were treated with compounds for 24 h. The cells were washed with PBS and incubated with 100 μl serum-free medium for 3 h. After incubation, cell culture medium was harvested. Cells were lysed with lysis buffer (50 mM Tris-HCL, 15 0mM NaCl, 1% (v/v) Triton X-100 with complete-Mini protease inhibitor cocktail) at 80 μl/well. The total protein concentration of cell lysate was quantified by Bradford assay. Clear flat bottom polystyrene high binding 96-well plates were pre-coated with goat anti-human AAT polyclonal antibody (80A, diluted AAT polyclonal antibody 1: 1000 in coating buffer (3.03 g Na₂CO₃, 6.0 g NaHCO₃ in 1L H₂O, pH = 9.6)) for 8–12 h at 4°C. The plate was washed 3x with PBST washing buffer (1xPBS with 0.1% Tween-20). 20 μl culture medium or cell lysate from each well was added to the pre-coated plate and the plate was incubated for 8–12 h at 4°C. The plate was washed 3x with PBST washing buffer. AAT monomer (16F8 1: 2000 in 1xPBS) or polymer (2C1 1:1000 in 1xPBS) specific antibody was added for incubation for 2 h at room temperature. Cells were washed 3x, followed by a HRP secondary conjugated goat-anti-mouse antibody (1:5000 in 1xPBS) and incubation for 2 h at room temperature. Following 3x wash, TMB substrate (Thermo Fisher Scientific, No. 34029) was added for 10 min and 2 M H₂SO₄ were used to stop the reaction. Plate was read by a BioTek Synergy H1 Hybrid plate reader at 450nm absorbance. AAT variant monomer levels were normalized to AAT-WT monomer level. AAT variant polymer level were normalized to AAT-Z polymer level.

AAT secreted protein NE inhibitory activity assay.

Cell culture medium was collected as described above. Culture medium was added to pre-coated a high binding 96-well plate. Plate was incubated in 4°C for 8–12 h. Following 3x wash, the plate was incubated with human NE at 5n g/well for 2 h at 37°C. Then 25 pmol/well elastase florescence substrate 2Rh110 was added and incubated for 1.5 h at 37°C. The plate was read by a BioTek Synergy H1 Hybrid plate reader at 485nm excitation and 525nm emission. The NE inhibitory activity for each variant was normalized to AAT-WT NE inhibitory activity.

Cycloheximide chase assay.

Huh7.5^null cell were transfected with AAT-Z DNA. After 24 h transfection, cells were treated compounds for 24 h. Then cells were treated with cycloheximide (CHX) (50 μM) for 0 h, 1 h, 2 h and 4 h. Following the CHX treatment chase, cell lysates were harvested and used for AAT polymer specific ELISA assay as we described above.

Variation spatial profiling (VSP) to build phenotype landscape.

The overall workflow of VSP is illustrated in the Fig. S4A. The VSP analysis was performed using gstat package (V2.0) in R. VSP is built on GP based machine learning. A special form of GP that is commonly used in Geostatistics, Ordinary Kriging, is used to model the spatial dependency as a variogram to interpolate the unmeasured value to construct the phenotype landscape for AAT. To generate a ‘phenotype landscape’, AAT variants were positioned by their sequence positions in the polypeptide chain on the ‘x’ axis coordinate with their impact on a phenotype on the ‘y’ axis coordinate compared to their impact on a second phenotype on the ‘z’ axis coordinate. Suppose the i^th (or j^th) observation in a dataset consists of a value z_i (or z_j) at coordinates x_i (or x_j) and y_i (or y_j). The distance h between the i^th and j^th observation is calculated by:

$$h_{(i,j)}=\sqrt{{\left(x_i-x_j\right)}^2+{\left(y_i-y_j\right)}^2}$$ (1)

The γ(h)-variance for a given distance (h) is defined by:

$$\gamma (h)=\frac{1}{2}{\left(z_i-z_j\right)}^2$$ (2)

where (h)-variance is the semivariance (i.e., the degree of dissimilarity) of the z value between the two observations, which is also the whole variance of z value for one observation at the given separation distance h, referred to as spatial variance here. The distance (h) and spatial variance (γ(h)) for all the data pairs are generated by the equations (1) and (2). Then, the average values of spatial variance for each distance interval are calculated to plot the averaged spatial variance versus distance. The fitting of variograms were determined using GS+ Version 10 (Gamma Design Software) by both minimizing the residual sum of squares (RSS) and maximizing the leave-one-out cross-validation result (see below). The variogram enables us to compute the spatial covariance (SCV) matrices for any possible separation vector. The SCV at the distance (h) is calculated by C(h)= C(0) − γ(h), where C(0) is the covariance at zero distance representing the global variance of the data points under consideration (i.e., the plateau of the variogram). The approach aims to generate the prediction that has minimized estimation error, i.e., error variance, which is generated according to the expression:

$${\sigma }_u^2=E[{\left(z_u{}^{*}-z_u\right)}^2]={\sum }_{i=1}^n{\sum }_{j=1}^n{\omega }_i{\omega }_j{C}_{i,j}-2\sum _{i=1}^n{\omega }_i{C}_{i,u}+C_{u,u}$$ (3)

where z_𝑢∗ is the prediction value while z_u is the true but unknown value, C_i,j and C_i,u are SCV between data points i and j, and data points i and u, respectively, and C_u,u is the SCV within location u. ω_i is the weight for data point i.

The SCV is obtained from the above molecular variogram analysis and the weight (ω_i) solved from equation (3) is used for following prediction. To ensure an unbiased result, the sum of weight is set as one:

$${\sum }_{i=1}^n{\omega }_i=1$$ (4)

Equations (3) and (4) not only solved the set of weights associated with input observations, but also provide the minimized ‘molecular variance’ at location u which can be expressed as:

$${\sigma }_u^2=C_{u,u}-\left({\sum }_{i=1}^n{\omega }_i{C}_{i,u}+\mathrm{\mu }\right)$$ (5)

where C_u,u is the SCV within location u, ω_i is the weight for data point i, and C_i,u are SCV between data points i and u. μ is the Lagrange Parameter that is used to convert the constrained minimization problem in Equation (5) into an unconstrained one. The resulting minimized molecular variance assessing the prediction uncertainty presents the confidence level of the prediction.

With the solved weights, we can calculate the prediction of all unknown values to generate the complete fitness landscape by the equation:

$${\text{z}}_u{}^{*}={\sum }_{i=1}^n{\omega }_i{\text{z}}_i$$ (6)

where z_𝑢∗ is the prediction value for the unknown data point u, ω_i is the weight for the known data point, and z_i is the measured value for data point i.

Validation of VSP modeling.

Leave-one-out cross-validation (LOOCV) is the best method that can be used to validate our computational models given small sample size. In the LOOCV, we remove each data point, one at a time and use the rest of the data points to predict the missing value. We repeat the prediction for all data points and compare the prediction results to the measured value to generate the Pearson’s r-value and its associated p-value. To further validate the model, we used a variant, K283I, that was not used as input for the computational modeling. This variant is located on the alpha-helix hG of AAT 3D structure (Fig. S7A). At the basal condition without drug treatment, K283I has a high intracellular polymer level (Fig. S7B) and ~36% of the NE inhibitory activity of WT (Fig. S7C), similar to AAT-Z. AA-147 does not significantly impact the intracellular polymer level of this variant (Fig. S7B), but significantly increases its NE inhibitory activity (by ~43%) (Fig. S7C). Similar to the measured value, the predictions from the SCV phenotype landscapes show no significant impact of AA-147 on the intracellular polymer level of this variant (Fig. S7D), but indicate around ~35% increase of NE inhibitory activity (Fig. S7E). These results suggests that our SCV approach can generate an accurate prediction of a variant for which the computational model was not trained, consistent with the strong correlation between predicted values and actual measurements in the leave-one-out cross-validation results.

Defining the high confidence prediction using probabilistic clustering.

Each prediction on the phenotype landscape is associated with a GP generated variance to indicate the prediction confidence. To separate the high confidence vs low confidence predictions in the GP generated phenotype landscape, we fit a Gaussian mixture model with two components over the GP generated variance by using mclust package (V.5.4.10) in R. Gaussian mixture model is a probabilistic clustering tool that not only separates the predictions with low variance vs high variance, but also outputs the probability distributions for them. We use the mean of the distribution for the low variance as a cut off and define the predictions with a lower variance as the high confidence prediction. The mean and the standard deviation of the distribution for the low variance (i.e., high confidence) are illustrated as contours in the phenotype landscape.

Inverse variance weighting (IVW) to build the residue-by-residue phenotype barcode.

Phenotype landscapes built based on a sparse collection of input variants map the full range of values describing function (based on the y- and z-axis metrics) for the entire polypeptide sequence (x-axis). To get an averaged value of predicted phenotype for each residue, we use the reciprocal of GP generated variance for the high confidence predictions as weights to aggregate the phenotype values by using the following equation:

$$\widehat{z}=\frac{{\sum }_i\frac{Z_i}{{\sigma }_i^2}}{{\sum }_i\frac{1}{{\sigma }_i^2}}$$ (7)

where $\widehat{z}$ is the weighted mean value for each residue, 𝑧_𝑖 is predicted phenotype value at z-axis for every value on the y-axis, ${\sigma }_i^2$ is the GP generated variance for each prediction. We repeat this process for all the residues. The IVW averaged mean values for all the residues are used to define barcodes showing the expected functional value for each residue in the polypeptide sequence and are mapped on AAT structures (PDB:3NE4 for AAT monomer, PDB:2D26 for the AAT-elastase complex and PDB:3T1P for polymer). Atomic resolution structures produced with PyMOL.

QUANTIFICATION AND STATISTICAL ANALYSIS

Experimental data are shown as the mean ± SD from 3 or more than 3 independent experiments. The differences between two independent groups were analyzed using Student’s t test. A P-value < 0.05 was considered statistically significant. The p-value of the Pearson’s r-value is performed using ANOVA test. The null hypothesis for the p-value of the Pearson correlation is that the correlation coefficient is not significantly different from 0. The p-value represents the probability that the correlation occurred by chance. We used p<0.05 to reject the null hypothesis and conclude there is a significant Pearson correlation.

Highlights:

• ATF6 activators improve monomer secretion and activity for pathogenic AAT variants

• ATF6 activators reduce polymer accumulation for pathogenic AAT variants including Z

• Gaussian process (GP) captures the critical role of gate area for AAT specific activity

• GP describes residue-by-residue conversion of the fold in response to proteostasis

Significance:

Genetic variation in the human population is responsible for protein misfolding and aggregation leading to proteotoxic stress in the cell associated with both systemic and neurodegenerative disorders. Pharmacological modulation of unfolded protein response (UPR) that regulates the stoichiometry of proteostasis network components provides a promising strategy for disease intervention. Alpha-1 Antitrypsin Deficiency (AATD) is caused by genetic variation in AAT through liver aggregation-associated gain-of-toxic pathology in the endoplasmic reticulum (ER) and insufficient neutrophil elastase (NE) inhibitory activity in the lung manifesting as chronic obstructive pulmonary disease (COPD). By integrating the impact of natural genetic variant information driving AATD using a Gaussian process (GP) based machine learning approach, we capture in an unprecedented way the residue-by-residue conversion of the pathogenic AAT fold into a functional protein by the enhanced proteostasis environment through pharmacological activation of the ATF6 branch of UPR. The conversion of the fold by proteostasis management is not only observed in terms of polymer/aggregation dissolution, but also, and unexpectedly, is able to increase the specific activity of the rescued extracellular monomeric AAT for reaction with NE, suggesting a novel strategy to address both the liver and lung disease phenotypes for AATD patients. These results, spanning inter-organ communication pathways, highlight the unsuspected role of proteostasis in the ER in (re)designing the extracellular properties of the fold for downstream function. Our residue-by-residue mechanistic view of the convertible state of the fold profiled through GP-based machine learning establishes a new spatial covariant (SCV) paradigm for understanding of the in vivo sequence-to-function-to-structure relationships that are managed by proteostasis. The GP-based SCV paradigm embraces the emergent role of proteostasis as an integral part of the protein folding design to control protein function during natural selection in response to the environment, which can be harnessed intelligently for precision management of human health and diseases.