Structural basis of CDNF interaction with the UPR regulator GRP78

Abstract

Cerebral dopamine neurotrophic factor (CDNF) is an unconventional neurotrophic factor that is a disease-modifying drug candidate for Parkinson’s disease. CDNF has pleiotropic protective effects on stressed cells, but its mechanism of action remains incompletely understood. Here, we use state-of-the-art advanced structural techniques to resolve the structural basis of CDNF interaction with GRP78, the master regulator of the unfolded protein response (UPR) pathway. Subsequent binding studies confirm the obtained structural model of the complex, eventually revealing the interaction site of CDNF and GRP78. Finally, mutating the key residues of CDNF mediating its interaction with GRP78 not only results in impaired binding of CDNF but also abolishes the neuroprotective activity of CDNF-derived peptides in mesencephalic neuron cultures. These results suggest that the molecular interaction with GRP78 mediates the neuroprotective actions of CDNF and provide a structural basis for development of next generation CDNF-based therapeutic compounds against neurodegenerative diseases.

CDNF is a clinical trial candidate in Parkinson’s disease but the mechanism of action is not fully understood. Here, the authors use SAXS and NMR techniques to resolve the structure of CDNF in complex with GRP78 and show that this interaction is required for the neuroprotective action of CDNF.

Introduction

Cerebral dopamine neurotrophic factor (CDNF) and mesencephalic astrocyte-derived neurotrophic factor (MANF) form a family of unconventional neurotrophic factors and are considered potential disease-modifying agents for the treatment of neurodegenerative diseases, such as Parkinson’s disease^1–4. Both CDNF and MANF have shown robust neuroprotective and neurorestorative effects that have been associated with the modulation of endoplasmic reticulum (ER) stress and local inflammatory signaling^5–8. Both proteins are localized in the ER lumen and contain a KDEL-type ER retention signal but are also known to be secreted under certain conditions, such as ER stress^9–12.

ER stress is known to be the result of diverse physiological and pathological events^13–15. Activation of the unfolded protein response (UPR) pathway serves as the main homeostatic cellular response mechanism to ER stress¹³. The UPR pathway is composed of three parallel signaling arms, inositol requiring enzyme 1 (IRE1α), PKR-like ER kinase (PERK) and activating transcription factor 6 (ATF6). These three UPR pathways regulate various cellular targets ranging from translation initiation and transcriptional control to calcium homeostasis and ER-mitochondria crosstalk with the aim to quickly restore ER homeostasis^13,16. If this cytoprotective mode of action fails to resolve ER stress, UPR signaling can trigger apoptosis to eliminate dysfunctional cells¹⁷. Deregulated UPR and chronic ER stress have been linked to a variety of chronic diseases, including neurodegenerative diseases¹⁵, and modulation of UPR signaling has become an attractive therapeutic target^14,18.

Previous studies have indicated that both CDNF and MANF regulate UPR signaling^5,19. Both in vitro and in vivo studies indicated that the neuroprotective effects of CDNF are strongly linked to alleviation of ER stress via altered UPR signaling⁶. CDNF shares significant structural homology with MANF^20,21. Both CDNF and MANF have been reported to bind to the nucleotide-binding domain of GRP78 (GRP78-NBD), major ER chaperone protein^6,22. The crystal structure of MANF in complex with the GRP78-NBD has already been solved²². Interestingly, the neuroprotective activity of MANF appears to be linked to its interaction with IRE1α rather than GRP78^7,23. Interactome screening suggested that CDNF binds to several ER chaperones and requires UPR signaling to promote neuronal survival⁶.

Here, by using small-angle X-ray scattering (SAXS) methods and nuclear magnetic resonance (NMR) techniques we show that CDNF forms a complex with the regulatory, nucleotide-binding domain of GRP78 and the neuroprotective activity of CDNF is lost if this interaction is blocked. These findings shed more light on how CDNF protects cells from ER stress and create a structural basis for generating compounds that mimic the neuroprotective activities of CDNF.

Results

Structural model of GRP78-NBD:CDNF complex from SAXS and hybrid modeling

Our initial solution scattering data indicated CDNF formed a complex with the nucleotide-binding domain of GRP78 (GRP78-NBD; Fig. 1A). Next, we investigated the solution structure of the CDNF:GRP78-NBD complex using small-angle X-ray scattering (SAXS), a powerful technique to provide insights into the three-dimensional arrangement of the interacting molecules in solution. Evaluation of the experimental SAXS data collected in batch mode from the nucleotide-binding regulatory domain of GRP78 (GRP78-NBD) strongly suggested the presence of a monomer-dimer equilibrium in solution. Thus, the direct coupling of size exclusion chromatography to SAXS was used (SEC-SAXS) for the removal of dimers (Fig. 1A, pink trace). The parameters derived from the obtained scattering curve including radius of gyration (R_g), maximum distance (D_max) and molecular weight (MW) estimates were consistent with the expected values for monomeric GRP78-NBD (Table 1). Furthermore, the experimental data aligned remarkably well with the scattering computed by the program CRYSOL²⁴ from the crystal structure of the monomer of GRP78-NBD (3IUC.pdb²⁵), as illustrated in Fig. 2A (pink trace). When CDNF was introduced to GRP78-NBD, a pronounced shift in the SEC-SAXS elution peak towards an earlier elution volume was observed in comparison to the elution peak of pure GRP78-NBD (Fig. 1A, orange trace). This shift indicated an increase in MW supporting the notion that CDNF formed a complex with GRP78-NBD. To ensure complete binding of CDNF to GRP78-NBD, an excess amount of CDNF was added and the shoulder to the right of the main elution peak was observed indicating a successful removal of unbound CDNF (Fig. 1A). The analysis of the obtained scattering profile from the main peak of the GRP78-NBD:CDNF complex (orange curve) revealed systematic differences compared to the scattering data of pure GRP78-NBD protein (pink curve) especially in the very low s ranges as shown in Fig. 1B and summarized in Table 1. As the log I(s) vs s plots reflect reciprocal space, these small differences translate into differences in the overall size of the sample. Guinier analysis of these differences revealed a notable increase in the R_g values, which shifted from 2.2 nm (GRP78-NBD) to 2.8 nm (GRP78-NBD:CDNF). Additionally, analysis of the distance probability function p(r) (Fig. 1C) enables the extraction of the maximum distance (Dmax). This showed an increase from 6.5 nm (GRP78-NBD) to 10 nm, which is to be expected in case of a complex formation. The comparison of the ab initio models (Fig. 1D) generated from GRP78-NBD by itself (pink) and in complex with CDNF (orange) suggest additional volume in the envelope of the complex that may comprise the bound CDNF molecule. This observation strengthens the hypothesis that CDNF binds to GRP78-NBD and contributes to structural changes in the protein. In conclusion, comprehensive analysis of the experimental SEC-SAXS data reinforces the existence of a monomer-dimer equilibrium in GRP78-NBD and supports the notion of CDNF binding to GRP78-NBD.

Fig. 1. Comparison of solution scattering of three samples: GRP78-NBD alone and in complex with CDNF or C-CDNF.

GRP78-NBD alone is shown in pink, GRP78-NBD in complex with CDNF in orange and GRP78-NBD in complex with C-CDNF in green. A Elution profiles from SEC-SAXS runs of the three samples. Integrated SAXS intensities of unsubtracted data are plotted vs. frame number. B Final SAXS scattering profiles (plotted as log I(s) vs s where s = 4 πsinθ/λ) of the 3 samples (C) Distance probability functions, p(r) of the three samples. D Ab initio models (generated with Dammif) are presented as spheres. Source data are provided as a Source Data file.

Table 1.

SAXS sample details, data collection, analysis, and 3D modeling details

(a) Sample details and set-up
Organism	Human
Source (Catalog No. or reference)	E. coli
Buffer	PBS, pH 7.2, 1.5% glycerol
SAXS data collection mode	Size exclusion chromatography
Sample temperature (°C)	20 °C
In beam sample celle	1 mm quartz capillary
Sample concentration, mg/ml	~10.0
Sample injection volume, ml	10 μl
SEC column type	Superdex 200 5/100 inc
SEC flowrate, ml/min	0.35 ml/min
(b) SAS data collection
Data acquisition/reduction software	SASFLOW / Chromixs
Source/instrument description	EMBL P12, 6 M Pilatus
Measured s-range (smin – smax; nm-1)	0.05 – 4.4
Exposure time (s)	0.5 s (~ 40 frames averaged)
(c) SAS-derived structural parameters (from ATSAS)
	GRP78-NBD	GRP78-NBD: C-CDNF	GRP78-NBD: CDNF
Guinier Analysis
Rg ± σ (nm)	2.2 ± 0.1	2.4 ± 0.1	2.8 ± 0.1
data point range	20 – 182	9 – 169	3 – 131
Linear fit assessment (fidelity)	0.9	0.98	0.6
PDDF/P(r) analysis
Dmax (nm)	6.5 ± 0.2	6.9 ± 0.2	10 ± 0.5
P(r) fit assessment (fidelity)	0.89	0.9	0.79
Molecular weight estimates (kDa)
From chemical composition	42.0	49.0	63.0
From SAS	45 ± 3	50 ± 5	60 ± 5
Ab initio modeling
Software	Dammif	Dammif	Dammif
Envelope volume (nm3)	83	85	124
Ambiquity (Ambimeter, a-score)	0.7	1.4	0.5
Atomistic modeling methods
Software	Crysol	Crysol	Crysol
X2	1.28	1.1	1.2
(d) Data and model deposition
SASBDB IDs	SASDNW9	SASDQS6	SASDNX9

Fig. 2. Hybrid modeling of GRP78-NBD:CDNF and GRP78-NBD:C-CDNF complexes.

GRP78-NBD:CDNF complex is shown in orange, GRP78:C-CDNF in green and GRP78-NBD alone in pink. A SAXS profiles and respective fits. The atomistic modeling relies on the reduced χ² error-weighted scoring function implemented in Crysol²⁴. Curves as shown in Fig. 1A have been shifted along y-axis for better visibility. GRP78-NBD, in light pink fitted with monomeric state of GRP78 (3iuc.pdb), χ² = 1.28. Fit of GRP78-NBD:CDNF in orange with complex as shown in (B), χ² = 1.2. Fit of GRP78-NBD:C-CDNF as shown in (E), χ² = 1.1 in green. B Cartoon representation of hybrid model indicating the binding of CDNF to GRP78. C Overlay of hybrid model with ab initio reconstruction as shown in Fig. 1D. D Scattering curve of GRP78-NBD:C-CDNF shown in Kratky representation (I*s² vs s) emphasizing small differences between fit of just GRP78-NBD (pink) and GRP78-NBD:C-CDNF (green). E Cartoon representation of hybrid model indicating the binding of C-CDNF to GRP78. F Overlay of hybrid model with ab initio reconstruction as shown in Fig. 1D. Source data are provided as a Source Data file.

To elucidate the potential binding site between CDNF and GRP78-NBD, we conducted a multifaceted modeling approach with parallel strategies. First, rigid body modeling was performed using the program SASREF. The utilized algorithm iteratively refines the positions and orientations of the rigid bodies – here, GRP78-NBD and CDNF—within the complex. This optimization process aims to minimize discrepancies between the theoretical and experimental SAXS profiles by adjusting the relative arrangements of the rigid bodies. The output is given as χ² values. With this approach, very good models were obtained, with the best result obtaining a χ² value of 1.2 (Supplementary Fig. 1A). In general, this allows for a good approximation of the binding region, however, it is not optimized for forming energetically favorable binding interactions.

Thus, two other approaches were performed with homology modeling and a computational docking analysis using the pyDockSAXS program. In this analysis, a set of 10,000 models of the complex was generated and assessed. Thereby, the models were ranked according to their agreement with the experimental SAXS data (χ² values) as well as their energy profiles.

An overlay of twenty top-ranked models illustrating their potential binding configurations is presented in Supplementary Fig. 2 and Supplementary Fig. 3. All except one model predicted a similar binding site which coincides with the binding area of the rigid body modeling. The fit of the ‘outlier’ model bound to the opposite side (#4672, Supplementary Fig. 3) demonstrates the issue of ambiguity in SAXS data and emphasizes the fact that deciphering complex shapes from scattering data alone is challenging as different models can produce similar scattering curves (Supplementary Fig. 4).

The other models docked to a similar region on GRP78-NBD, yet with different orientations of CDNF. In Supplementary Fig. 3, the models and fits of selected models are shown. The final χ² values of the top 20 binders were between 2.7 and 4.1. The selected models included the best binder with N-terminal CDNF docked (#81, ranked number 2 with χ² value of 2.7) and the best binder with the C-terminal domain of CDNF docked (#8027, ranked number 9 with χ² value of 3.2). The fits indicated that the real complex is slightly more compact than the current models—independent of the binding site and binding orientation of CDNF.

Compared to the rigid body modeling, similar binding regions were selected, however, the computational docking approach did not achieve the same degree in χ² fitting (Supplementary Fig. 3). Furthermore, literature search and homology modeling suggested a potential shared binding site with the homolog MANF (6ha7.pdb²². Supplementary Fig. 1B), which is also in good agreement with the binding sites suggested by the rigid body modeling approach. Interestingly, the fit of the complex consisting of GRP78-NBD:MANF roughly fits the data with a χ² = 3.2 (Supplementary Fig. 1B). Indeed, the top 20 binders of the computational docking approach included two models with a good overlay with the MANF binding. Model #8583 (ranked # 16, with a χ² = 3.9) is shown in Supplementary Figs. 2 and 3 for comparison.

Based on this observation, we generated several homology models, replacing the MANF molecule and accounting for the observed flexibility of the linker between the C- and N-terminal domains of CDNF. We produced 20 hybrid models based on the crystal structure of GRP78-NBD and the solution NMR structure of CDNF (4 bit.pdb). For this, the C-terminal domain of CDNF was overlaid with MANF molecule (6ha7.pdb), and 20 individual models were produced reflecting the variable position on the N-terminal domain of CDNF and ranked them according to their χ² values obtained with CRYSOL. With this step, we identified a model (based on solution state 13) with an improved fit from χ² > 6 to χ² = 3.5 (Supplementary Figs. 1C and 2). In addition, we allowed for improved docking by employing Normal Mode Analysis (NMA) with the ATSAS program SREFLEX.

The results of the NMA analysis of all 20 hybrid models were assessed regarding the number of induced sterical clashes and bond breaks as well as the number of formed salt and hydrogen bridges. For the model based on solution state 13 a final χ² = 1.2 was achieved (Fig. 2A, B, Supplementary Fig. 1D, and Supplementary Fig. 2 D). In Fig. 2C, an overlay of this final hybrid model with the ab initio model is presented, showcasing their alignment. These data and the structural models have been deposited into SASBDB archive (see Methods).

GRP78-NBD structure in complex with the C-terminal domain of CDNF

Combining hybrid SAXS modeling with comparisons of the binding sites with known structures provided valuable clues regarding the potential location and mode of interaction between the C-terminal domain of CDNF and the GRP78-NBD. Similar to the full-length CDNF, the SEC-SAXS data obtained from the mixture of GRP78-NBD with C-CDNF indicated binding, as evidenced by the observed shift in the SEC elution peak (Fig. 1A, green trace). Here, the excess unbound C-CDNF is evident in a smaller elution peak following the main peak. Although the final GRP78-NBD:C-CDNF SAXS profile did not exhibit significant differences compared to that of pure GRP78-NBD (Fig. 1B), a closer examination of the data revealed slight distinctions in the low s range, which became more apparent in the p(r) analysis (Fig. 1C). The ab initio model generated from the SAXS data also showed a slightly larger envelope (see Table 1) in the presence of C-CDNF (Fig. 1D).

Using the earlier proposed model for the GRP78-NBD:CDNF complex, we generated a hybrid model consisting solely of the C-terminal domain of CDNF bound to GRP78-NBD (Fig. 2D–F). The theoretical scattering curve derived from this hybrid model was then compared to the experimental SAXS data (Fig. 2D). The fit between these two curves was excellent, as demonstrated by the low χ² value of 1.1 (Fig. 2A, green curve). In comparison, the unbound version of GRP78-NBD exhibited a somewhat worse χ² value of 1.64. The small discrepancy becomes more evident when examining the Kratky plot representation, which emphasizes the scattering intensities in the mid-angle region (Fig. 2D).

In summary, these structural analyses of the GRP78-NBD:CDNF and GRP78-NBD:C-CDNF complexes, support the notion that the binding interaction with GRP78-NBD occurs via the C-terminal domain of CDNF.

Identification of key residues in the GRP78-CDNF binding interface via orthogonal methods

Figure 3 highlights the binding interface between the C-terminal domain of CDNF and the cleft formed by GRP78-NBD’s Ia and IIa subdomains, similar to the binding mode observed for MANF and GRP78-NBD previously²². Figure 3 (also Supplementary Fig. 1C, D) highlights the fact that the same pocket of GRP78-NBD is involved in the interactions with MANF and CDNF.

Fig. 3. The binding between CDNF and GRP78-NBD is mediated by residues 156-173 of the C-terminal domain of CDNF.

A An overview of the complex is shown in cartoon in comparison to the MANF binding to GRP78-NBD. MANF is shown in green, CDNF in blue, and GRP78-NBD in light pink. B Enlarged view of the binding interaction site of CDNF to GRP78-NBD. Selected residues of the interface are shown in stick representation (marine blue in CDNF and dark pink in GRP78-NBD). C158 and C161 forming the di-sulfide bridge in CDNF is shown in yellow. Salt bridges are indicated. C The GRP78-NBD residues forming the binding interface are shown in stick representation and are highlighted in dark pink. D C-CDNF residues along the interaction site are shown in stick representation and are highlighted in marine blue. The disulfide bridge between C158 and C161 is shown in yellow. E C-CDNF WT sequence and mutations. C158 and C161 are marked in orange for orientation. Top line shows the wild-type (WT) sequence with the potential GRP78-NBD interacting residues (as indicated by the SAXS analysis) underlined. Lines 2–4 show the C-CDNF mutant sequences that were used in biophysical characterization and cell-based assays. The line on the bottom shows the labeled alanine residues that showed an effect in HSQC NMR titration (highlighted in cyan). Alanine residues that did not show an effect are shown in gray.

Whereas we used ITC to confirm binding between full-length CDNF and GRP78-NBD (Fig. 4A), we used spectral shift assay to analyse the binding of the smaller peptide C-CDNF and its mutants. To further explore the potential binding of GRP78-NBD to the C-terminal domain of CDNF, we conducted binding experiments using the C-terminal 60 residues (C-CDNF). In order to further investigate the role of the bulky side chains of the other non-Cys residues (E156, E157 and R159) in the GRP78-binding region of CDNF, we also synthesized a series of C-CDNF mutants (Fig. 4C): C-CDNF-MT1 (E163A/K164A), C-CDNF-MT2 (E156/157/163A), C-CDNF-MT3 (R159A/K164A).

Fig. 4. Binding studies of CDNF and C-CDNF variants to GRP78-NBD.

A ITC measurements allowing the detection of binding of CDNF to GRP78-NBD. The power differentials (DP) are plotted over time with the baseline indicated by a yellow line (left). Best-fit curve (black line) to the experimental data points (yellow)(right). The fit parameters for ITC: n(sites) = 0.346 ± 0.129; K_D(M) = 3.30 × 10^-6 ± 3.72 × 10^-6; ΔH(kcal/mol) = 7.48. B Spectral shift binding assay for C-CDNF and GRP78-NBD is shown in the scatter plot. The data of two independent measurements are shown in light blue and deep blue (left). The binding data with the C-CDNF mutants (altered residues highlighted in red) are summarized in the table (right) and are shown in detail in Supplementary Fig. 5. Source data are provided as a Source Data file.

These mutants were tested for GRP78-NBD binding (Fig. 4B) using spectral shift assays. The obtained data clearly indicates two mutants (C-CDNF-MT1 and C-CDNF-MT2) do not bind to GRP78-NBD. C-CDNF-MT3 showed impaired binding to GRP78-NBD compared to wild-type C-CDNF (K_D 2330 nM compared to K_D 380 nM with wild-type C-CDNF; Fig. 4B). This indicated that the residues E156, E157, E163 and K164 of CDNF have a crucial role in GRP78 binding.

The 2D¹H-¹³C HSQC NMR titration data (Fig. 5) with ¹³C/¹⁵N-Ala-labeled C-CDNF provide further support of C-CDNF binding to GRP78-NBD on an atomic level. The three well-resolved methyl groups of A145, A162 and A175 are clearly affected. A162 and A175 display essentially pure slow exchange behavior with the peak integrals proportional to the population of the free state of C-CDNF while A145 shows intermediate exchange (although close to the slow exchange regime on the NMR time scale). No emerging NMR peaks corresponding to the bound state of C-CDNF could be identified for these alanine residues at the present study conditions, presumably as the line broadening (R₂) of the peaks in this state is much higher than the exchange rate, i.e. the magnetization decay rate is faster than the build-up rate²⁶. Using the measured integral values of A162 and A175, a dissociation constant (K_D) in the interval of 0.6 to 7 μM was conservatively estimated as further outlined in the Supplementary table 3. In addition, the partially overlapped methyl group of A135 also appears to be affected by binding to GRP78-NBD with decaying intensity (slow exchange) while the two remaining labeled residues A179 and A180 do not display any clear effects in the titration experiment.

Fig. 5. Overlay of 2D ¹H-¹³C HSQC NMR methyl fingerprinting spectra for the various titration points.

The spectral data (chemical shifts) of methyl groups are diagnostic of the tertiary structure, Panel A shows the spectral region containing the ¹H−¹³C cross peaks of the methyl groups of the six ¹³C-labeled alanine residues in C-CDNF. Expansions of A162, A175 and the group of A135, A145, A179 and A180 are shown in panels B–D, respectively, showing the effects upon titration of GRP78-NBD (color codes: red = C-CDNF only, green = C-CDNF with 21 µM GRP78-NBD, teal = C-CDNF with 38 µM GRP78-NBD, blue = C-CDNF with 53 µM GRP78-NBD, purple = C-CDNF with 65 µM GRP78-NBD).

The observation of A145, A162 and A175 being most affected is consistent with the obtained SAXS and further binding data of GRP78-NBD and C-CDNF mutants. These residues also are located within or as part of secondary structure elements close to the proposed GRP78-NBD binding interface shown in Fig. 3.

C-CDNF mutants that do not bind to GRP78 fail to protect dopamine neurons from stress

In order to study the biological relevance of GRP78 binding of CDNF, we tested mutants of C-CDNF in an in vitro neuroprotection assay. In this model, rat primary mesencephalic neuron cultures are injured with MPP⁺, a mitochondrial toxin used to model dopaminergic neurodegeneration²⁷, and neuroprotective activity of compounds can be then tested²⁸. The key readouts in this assay are survival of TH-positive dopamine neurons and accumulation of α-synuclein (aSyn) aggregates in dopamine neurons after 48 h treatment with the MPP⁺ toxin, as determined by immunostaining and automated image analysis. While the wild-type C-CDNF (at 5 nM) increased dopamine neuron survival by 52.3 ± 7.4% (p = 0.0117) and reduced aSyn aggregation by 64.7 ± 9.1% (p < 0.0001) from the MPP⁺ model effect calculated as the difference between the control and MPP⁺ groups (p < 0.0001 vs cells treated with MPP⁺ only), C-CDNF-MT1 maximally increased dopamine neuron survival only by 13.6 ± 6.7% (p = 0.9771) with maximal reduction of aSyn aggregation by 21.2 ± 5.9% from MPP⁺ model effect (p = 0.4382 vs MPP⁺ treated cells) (Fig. 6A). Similarly, C-CDNF-MT2 showed very low activity in neuroprotection or aSyn aggregation at all three tested concentrations, compared to wild-type C-CDNF. C-CDNF-MT3 showed some neuroprotective activity (40.3 ± 8.4% increase from MPP⁺ model effect; p = 0.0878) and reduction in aSyn aggregation (−42.1 ± 7.7% from MPP⁺ model effect; p = 0.0073) at only the highest 500 nM concentration. Notably, C-CDNF-MT3, which had reduced potency in the neuroprotection assay, showed weak binding to GRP78-NBD while C-CDNF-MT1 and C-CDNF-MT2 (mutants that did not bind to GRP78-NBD) did not show any neuroprotective activity at any of the tested concentrations.

Fig. 6. Effects of C-CDNF and C-CDNF mutants on dopamine neuron survival, α-synuclein aggregation and UPR pathway signaling in a rat primary neuron MPP⁺ injury model.

A Effect of wild-type C-CDNF and C-CDNF mutants MT1-MT3 on survival of tyrosine hydroxylase (TH)-positive dopamine neurons (top) and α-synuclein (aSyn) aggregation (bottom) in TH-positive neurons in a primary culture of rat mesencephalic neurons after MPP⁺ injury. B Effect of wild-type C-CDNF and C-CDNF mutants MT1-MT3 on activation of the three UPR pathways IRE1α-XBP1 (top), PERK-ATF4 (middle) and ATF6 (bottom) in TH-positive neurons in a primary culture of rat mesencephalic neurons after MPP⁺ injury. The markers indicating activation of the UPR pathways were phosphorylated IRE1α (top), nucleus-localized ATF4 (middle), and nucleus-localized ATF6 (bottom). N = 4−6 per experimental condition. The results are presented as a mean percentage from the control level +/- SEM and are from separately run experiments (for A, B). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 vs MPP⁺ treated cells, ^#p < 0.0001 healthy vs MPP⁺ treated cells; one-way ANOVA with Dunnett’s test. Source data are provided as a Source Data file.

As GRP78 serves as the master regulator of the UPR pathway, binding to the ER luminal domains of the three UPR receptors IRE1α, PERK and ATF6, we also assessed UPR signaling activity in dopamine neurons in this model. All three UPR pathways were strongly activated by the MPP⁺ injury as shown by increased levels of S724 phosphorylated IRE1α (81.0 ± 5.0% increase from control cells; p < 0.0001), nucleus-localized, activated forms of ATF4 (an indicator of PERK activity; 42.1 ± 6.1, p < 0.0001 vs control cells) and ATF6 (79.5 ± 6.2, p < 0.0001 vs control cells) in TH⁺ dopamine neurons (Fig. 6B). C-CDNF strongly reduced the activity of all three markers indicating UPR activity. For example, IRE1α phosphorylation was reduced in dopamine neurons treated with 50 nM C-CDNF by 63.4 ± 12.8% from the MPP⁺ model effect (p = 0.0008) while the nucleus-localized forms of ATF4 and ATF6 were reduced by 57.9 ± 13.1% (p = 0.0484) and 91.7 ± 2.2% (p < 0.0001), respectively. Notably, C-CDNF-MT1 and C-CDNF-MT2 had nearly completely lost the UPR modulating effect on IRE1α and PERK/ATF4 pathways compared to wild-type C-CDNF. C-CDNF-MT3 showed a loss of UPR modulating activity at 5 and 50 nM but a UPR-modulating effect at 500 nM [67.2 ± 6.0% (p = 0.0003) decrease in phospho-IRE1α and 55.3 ± 8.2% decrease (p = 0.0498) in nuclear ATF4], which correlated well with the neuronal survival and aSyn readouts (Fig. 6A). Interestingly, all three C-CDNF mutants retained some ATF6 pathway modulating activity in the tested concentrations (5-500 nM). A dose-dependent effect was again seen with C-CDNF-MT3, which could indicate that residual binding to GRP78-NBD may explain activity at higher concentrations. It is possible that the C-CDNF effect on the ATF6 pathway could involve additional, GRP78-independent interactions between the molecules that currently remain unknown. Nevertheless, despite the remaining ATF6 activity in TH⁺ cells in the presence of C-CDNF mutants, the neuroprotective and aSyn aggregate-reducing effects were abolished suggesting that IRE1α and PERK pathways play a more central role in mediating the neuroprotective effects of C-CDNF in this model.

In conclusion, those C-CDNF mutants that failed to bind to GRP78-NBD could not rescue dopamine neurons from MPP⁺-induced neurodegeneration. The loss of UPR modulating (particularly IRE1α and PERK pathways) and proteostatic (aSyn aggregation inhibiting) activity in the C-CDNF mutants that cannot bind to GRP78 supports the concept of GRP78-UPR pathway serving as the key mediator of the neuroprotective activities of CDNF.

Discussion

As an ER luminal protein CDNF is well-positioned for modulation of ER stress response and UPR signaling. However, it has been poorly understood what molecular events mediate these effects. CDNF levels are regulated by ER stress and its neuroprotective activity is dependent on IRE1α and PERK signaling activity. While CDNF can protect cells from ER stress-induced apoptosis it does not seem to function as an antagonist of UPR signaling⁶. Molecular interactions in the ER lumen are likely part of this regulatory mechanism. The data presented here are in line with this and suggest that GRP78 interaction is the key molecular interaction required for the neuroprotective activity of CDNF, in line with a recent publication by the Saarma group⁶. Further studies are required to better understand the complex molecular choreography based on which CDNF modulates UPR activity via GRP78, the UPR receptors and other UPR regulating proteins.

CDNF shows structural homology to another ER stress-regulating protein MANF^20,21. The GRP78-NBD:MANF complex structure²² showed previously that MANF binds to the regulatory, nucleotide-binding domain of GRP78 via its C-terminal domain. The residues E153 and K154 of MANF, which are homologous to E163 and K164 of CDNF respectively, are also seen to be involved in the GRP78 binding interface. The data presented here clearly highlight the fact that CDNF binds to the nucleotide binding domain of GRP78 via its C-terminus. Similarities of the interaction of GRP78 with CDNF and MANF (as shown by the crystal structures 6H9U and 6HA7) are limited within the domain level interactions (nucleotide-binding domain of GRP78 and C-terminal domains of CDNF and MANF) only. We did not find the exact same set of homologous residues (of CDNF and MANF) being involved in the interaction (Fig. 3; Supplementary Fig. 1C, D) with GRP78 although clearly the area surrounding the CXXC motif appears to be critical for both proteins. Since the interactions (GRP78-NBD:MANF and GRP78-NBD:CDNF) were resolved using different structural methods (X-ray crystallography and SAXS) respectively, the resolution of the derived final models were different in nature. Interestingly, while MANF also regulates UPR signaling activity it was recently proposed that IRE1α rather than GRP78 would be the key molecular interaction for neuroprotective activity of MANF⁷. Future studies are expected to provide a more refined molecular-level understanding of similarities and differences in how CDNF and MANF regulate the UPR pathway and help cells to cope under ER stress.

When relying on structural information derived from SAXS, one must address the issue of ambiguity in SAXS data²⁹. Often the data can be fitted with various structural models, which can introduce uncertainty in determining the most accurate representation of the biological system under investigation. In this study, we used several tools to reduce ambiguity and carefully designed the data collection strategy, applying and comparing various modeling approaches, and, importantly, making use of additional information and thorough validation techniques to refine the models. During the process, one obvious ‘outlier’ was obtained with the computer-based docking approach, suggested binding of CDNF to the opposite region of GRP78-NBD. The overlay with the final model showed that despite the large difference at the atomic level the overall shape is somewhat similar and, thus, explains how it was able to be selected during the computer-based docking approach due to a relatively good χ² value.

Interpreting transient interactions and weak bindings remains challenging, as the samples are polydispersive in nature, and streamlined data analysis procedures cannot be applied. To address this issue, on-the-fly sample separation was achieved by directly coupling size-exclusion chromatography with SAXS. Additionally, collecting data on full-length CDNF and conducting a comprehensive analysis of the data increased confidence in our proposed model of the complex, as opposed to simply concentrating on the binding of the C-CDNF fragment. The SAXS data of the larger complex has an ambiguity score of 0.5²⁹, whereas the smaller construct is assigned a score of 1.4, just below the cutoff rate of 1.5 at which the SAXS data sets are flagged with the warning that caution is needed. Furthermore, our multifaceted modeling approaches all delivered a similar binding interface, independent of the initial method that was applied, including the use of molecular dynamics, homology modeling, rigid body modeling with and without constraints, and the inclusion of flexibility. All modeling attempts led to the identification of a similar binding site. This increased confidence in selecting our final model based on the one hand on the homology modeling with the known MANF bound complex and on the other hand on the solution structure of CDNF including the 20 different states displaying the flexibility of the linker between the N- and C-terminal domains.

This study demonstrates that SAXS is a powerful tool to combine various pieces of information from biochemical, biophysical and structural studies. The identified binding interface allowed for an interpretation of the role of CDNF in regulation of the UPR pathway. The biophysical studies along with mutation analysis of key residues in the CDNF-GRP78 binding interface strongly suggested that GRP78 is the key mediator of UPR modulation and neuroprotective effects of CDNF. The structural data presented here have proven useful for development of compounds mimicking CDNF action. Recently, a novel blood-brain barrier-penetrating CDNF mimic HER-096 was developed based on the GRP78-binding interface of CDNF³⁰. This compound is developed as a disease-modifying therapy for Parkinson’s disease and recently a first-in-human Phase 1a clinical study (ClinicalTrials.gov ID: NCT05915247) has been completed. It will be intriguing to see if HER-096 and other UPR-modulating compounds can eventually slow down disease progression in chronic neurodegenerative conditions, and potentially in other diseases in which deregulated UPR signaling or chronic ER stress play a central pathogenic role.

Methods

Cloning of GRP78-NBD and CDNF

The DNA sequences of CDNF_WT (aa 25-187) along with the DNA sequence of GRP-78-NBD (aa 26-382) were ordered as synthetic genes (Thermo Fisher Scientific, GeneArt). Synthetic genes were readily subcloned into pDONR221 vector (Thermo Fisher Scientific) that has LR- cloning compatible sites and thereafter cloned into pDest527 plasmid (GRP78-NBD, from Addgene, #11518) and pDest566 plasmid (CDNF, from Addgene, #11517) vector using LR Clonase II kit (Invitrogen) respectively. Both vectors have N-terminal HisTag followed by a TEV protease site before the gene of interest. pDest566 does have an additional maltose binding protein (MBP) tag in between the His tag and the TEV protease site. After cloning, the plasmids were transfected into TOP10 Escherichia coli cells (Thermo Fisher, Invitrogen, UK) with heat shock method and single colonies from plates were picked and cultivated in LB_amp to isolate plasmids with QIAprep Spin Miniprep Kit (Qiagen). DNAs coding protein of interest were confirmed by sequencing plasmids with T7 forward primer (5’-TAATACGACTCACTATAGGG-3’) in case of GRP78-NBD (pDest527) and T7 MBP forward primer (5’-GATGAAGCCCTGAAAGACGCGCAG-3’) in case of CDNF (pDest566). T7 terminal primer (5’-GCTAGTTATTGCTCAGCGG-3’) was used as a reverse primer for both proteins.

Expression of CDNF and GRP78-NBD

Wild-type CDNF was expressed using T7 Shuffle Express E. coli strain (New England Biolabs) to retain the disulfide bridges in them. Plasmids were transformed into the T7 Shuffle Express cells with the heat shock method and single colonies were used to make glycerol stocks. These were used to inoculate 10 ml LB_amp+glucose medium which were then used to initiate batch mode cultivations of 500 ml of LB_amp medium in incubator shaker at +37 °C. Expression was induced with 0.1 mM IPTG after the OD₆₀₀ reached 0.5–0.7. E. coli cultivations were continued for an additional 18 h at +20 °C before the cells were harvested by centrifugation. The pellets were washed once with TBS, pH 7.4 before being used for protein isolation.

GRP78-NBD plasmid was transfected into the E. coli expression strain BL21 Star™ (DE3) One Shot® using the heat shock method. A single colony from the LB_amp+glucose plate was used to make glycerol stocks which were used to inoculate 10 ml LB_amp+glucose medium that was eventually used to inoculate 500 ml LB_amp expression culture medium in an incubator shaker at +37 °C. The induction of the gene expression was performed by adding 0.1 mM IPTG after the optical density at 600 nm (OD₆₀₀) reached the range of 0.4 to 0.6. After an incubation time of 20 h at +20 °C the cells were harvested by centrifugation and washed once with 50 mM HEPES, 200 mM NaCl, 10% glycerol, pH 7.5 before being used for protein isolation.

Purification of CDNF and GRP78-NBD

CDNF was purified with affinity chromatography with a N-terminal HisTag as the purification tag. Before the chromatography step, protein isolation began by dissolving the cell pellets in lysis buffer containing 50 mM NaPO₄, 500 mM NaCl, pH 7.4, 50 µg/ml lysozyme and Complete Protease Inhibitor Cocktail tablet (Roche, #11697498001). The cells were then lysed with 5 min sonication (40%, 5 s on, 1 s off). After centrifugation to remove the cell debris, the clarified supernatant was applied to HisPur Ni-NTA agarose resin (Pierce #88222) in batch mode overnight at 4 °C. Impurities were washed away with 10 column volumes of 50 mM NaPO₄, 500 mM NaCl, pH 7.4 buffer, followed by the elution of the bound protein with six column volumes of elution buffer (50 mM NaPO₄, 500 mM NaCl, 250 mM imidazole, pH 7.4). The eluent was collected into a single fraction to which 1 mM of EDTA was added to prevent precipitation due to leaked Ni-ions from the column. The protein concentrations of the CDNF constructs were measured at λ = 280 nm with a NanoDropOne instrument using extinction coefficients ε = 84270 M^-1 cm^-1, calculated from the amino acid sequences. The quality of the protein was analyzed with SDS-PAGE (Supplementary Fig. 6).

Protein isolation of the expressed GRP78-NBD gene was carried on with immobilized cobalt metal affinity chromatography. Here, cell lysis was performed in 50 mM HEPES, 200 mM NaCl, 10 % Glycerol, pH 7.5, 5 mM ATP and 10 mM MgCl₂ buffer and using an Emulsiflex C3 high-pressure homogenizer (Avestin Inc., Ottawa, Canada). After four iterations, the cell debris were removed by centrifugation and the clarified supernatant was incubated with Talon Superflow resin (GE Healthcare, #28-9575-02) in batch mode overnight at 4 °C, respectively. The bound proteins were washed with 12 column volumes of 50 mM HEPES, 200 mM NaCl, 10 % Glycerol, pH 7.5, 5 mM ATP and 10 mM MgCl₂ buffer followed by the elution with five column volumes elution buffer (50 mM HEPES, 200 mM NaCl, 10% Glycerol, 250 mM imidazole, pH 7.5) and the eluent was collected into a single fraction. The protein concentration was measured at λ = 280 nm with a NanoDropOne instrument using extinction coefficient ε = 21890 M⁻¹ cm⁻¹ calculated from the amino acid sequence using Protparam³¹. The quality of the protein was analyzed with SDS-PAGE (Supplementary Fig. 7).

Removal of affinity tags from the proteins

The purification tags were removed using TEV protease, after which buffer of the proteins were changed with dialysis to 50 mM NaPO₄, 500 mM NaCl, pH 7.4, in a molar ratio of 30:1 during an overnight reaction at +4 °C. Un-cleaved protein and traces of TEV proteases were removed from solution using Ni-NTA affinity chromatography and proteins without purification tags were recovered from the flow through and concentrated using a 10 K MWCO centrifugal filters (Sartorius, #VS15T01).

To improve the purity of the CDNF and GRP78-NBD, these were additionally passed through a Superdex Increase 75 PG column after tag removal and fractionated (Supplementary Fig. 6B). Protein containing fractions were pooled together and concentrated using a 10 K MWCO centrifugal filters (Sartorius, #VS15T01). As a final quality check the degree of polydispersity of the freshly purified proteins were determined with dynamic laser scattering (DLS) using a Zetasizer Nano ZS instrument (Malvern Instruments Ltd.). The hydrodynamic diameter of the proteins were determined as the average of three measurements (each measurement containing 10–20 10 s datasets at 25 °C; Supplementary Figs. 6C and 7B).

SAXS data collection

Synchrotron SAXS data were collected on the EMBL P12 beamline at PETRA III (DESY, Hamburg, Germany³²) using a Pilatus 6 M detector at a sample-detector distance of 3 m and at a wavelength of λ = 0.124 nm. The data underwent standard automated processing (radial averaging, background subtraction, etc.) combined with manual evaluation to produce 1D SAXS profiles which are given as (I(s) vs s, where s = 4πsinθ/λ, and 2θ is the scattering angle).

The data were collected at 20 °C using the standard batch mode set up with automated sample delivery. Samples and the corresponding matched solvent blank were measured under continuous sample flow using 40 × 0.1 s exposure times. Generally, the individual samples were measured at several dilutions in the concentration range from 0.25 – 10 mg/ml. All measurements were performed in PBS pH 7.2.

For a thorough characterization of GRP78-NBD and the complexes thereof, in-line size-exclusion chromatography (SEC) SAXS was employed³³. The SEC parameters were as follows: 10 μl sample at 10 mg/ml was injected at a flow rate of 0.35 ml/min onto a GE Superdex 200 Increase 5/150 column at 20 °C. 1800 successive frames, each with 0.5 s exposure times, were collected. The buffer was supplemented with 1.5% glycerol to reduce radiation damage. The data were normalized to the intensity of the transmitted beam and radially averaged; the scattering of the solvent-blank was subtracted. The elution peak comprised 30 frames. SEC-SAXS runs were performed on GRP78-NBD, GRP78-NBD:CDNF and GRP78-NBD:C-CDNF.

Initial SAXS data processing was performed with SASFLOW from the ATSAS 3.1 package³⁴, the automated analysis pipeline at the P12 beamline and Chromixs was used for SEC-SAXS data analysis³⁵. The data were re-processed manually to produce the final working scattering profiles.

SAXS data modeling

For ab initio modeling the ATSAS program DAMMIF was employed³⁶. Here, the results of 10 individual rounds of modeling are compared to each other to assess the ambiguity of the data (DAMAVER³⁷). Several alternative approaches were utilized to generate reconstructions based on the SAXS data. A rigid body modeling was conducted with the programs SASREF/SASREFMX³⁸ as well as CORAL³⁹ as described in the Results in more detail. In addition, the program pyDockSAXS⁴⁰ was used to MD simulate a large pool (10.000) of possible binding sites for CDNF to GRP78-NBD. These were then ranked according to their fit against the experimental scattering data.

In addition, homology models were also generated based on the MANF:GRP78-NBD complex (6ha7.pdb²²). For this, the program PYMOL was used for overlaying the respective molecules and generating ‘hybrids’. Before continuing with ongoing processing steps, the hydrogens were removed, to reduce complications with only partially hydrated atomic coordinates. Further refinement of the docked CDNF was achieved through SREFLEX⁴¹, leveraging Normal Mode Analysis (NMA) to enable subtle local adjustments. Assessment of the generated models involved evaluating steric clashes, and bond breakages, as well as identifying resulting hydrogen bonds and salt bridges. The latter was facilitated by the EBI PISA server⁴².

Goodness of fit of all the respective models was determined with the program CRYSOL³⁸ as implemented in ATSAS (version 3.05).

Solid state synthesis of C-CDNF peptides

The peptides were synthesized in two parts, an N-terminal and a C-terminal part followed by native chemical Ligation (NCL) forming the A160-C161 peptide bond.

Synthesis of N-terminal part of C-CDNF (127-160 aa): Fmoc-Cys(StBu)Gly-N-2-hydroxybenzyl preloaded resin (Hnb-resin) was prepared according to a literature procedure⁴³ in on a TentaGel S RAM resin (loading 0.2 mmol/g). The first amino acid Ala¹⁶⁰ was loaded using Fmoc-Ala-OH (10 eq, 0.4 M in DMF), HCTU (10 eq., 0.5 M in DMF) and DIEA (20 eq, 0.5 M in DMF), and the coupling was performed by shaking overnight at room temperature (RT). The unreacted secondary amine was capped with Ac₂O (20% in DMF) and DIEA by standard method. All the remaining amino acids were coupled on an automated peptide synthesized, Biotage Initiator+ Alstra or Activotec Activo-P11 instrument. Fmoc groups were removed with 20% piperidine in DMF and all couplings were performed using amino acid (4 eq, 0.4 M in DMF), HBTU (4 eq., 0.5 M in DMF), and DIEA (4 eq, 0.5 M in DMF) at RT for 60 min. After the complete chain assembly, the resin was washed thoroughly with methanol and dried in a desiccator. The peptide was released from the resin using TFA/TIS/Water (95:2:3, 20 mL) for 2 h. The resin was filtered off, excess TFA was evaporated, and the peptide was precipitated in ice-cold Et₂O, collected, dissolved in a mixture of Acetonitrile (MeCN) and water and lyophilized to obtain the crude product. The crude peptide was purified using RP-HPLC to yield the pure N-terminal part of C-CDNF (127-160)-Hnb.

Synthesis of the C-terminal part of C-CDNF (161-187 aa): C-terminal part of C-CDNF (161–187 aa) was synthesized using standard Fmoc solid-phase synthesis, starting with commercially available Fmoc-Leucine pre-loaded TentaGel PHB resin. Amino acids were coupled on an automated peptide synthesizer, Biotage Initiator+ Alstra or Activotec Activo-P11, Fmoc groups were removed with 20% piperidine in DMF and all couplings were performed using amino acid (4 eq, 0.4 M in DMF), Oxyma-K (4 eq., 0.5 M in DMF), and DIC (4 eq, 0.5 M in DMF) for 1 h at RT. After the complete chain assembly, the resin was washed thoroughly with methanol and dried in a desiccator. The peptide was released from the resin using TFA/TIS/Water (95:2:3) for 2 h. The resin was filtered off, excess TFA was evaporated, and the peptide was precipitated in ice-cold Et₂O, collected, dissolved in a mixture of MeCN and water and lyophilized to obtain the crude product. The crude peptide was purified using RP-HPLC to yield the pure C-terminal C-CDNF (161-187).

Native Chemical Ligation (NCL) and disulfide bond formation of C-CDNF (127-187 aa): N-terminal part of C-CDNF (127-160)-Hnb (1) (1 eq) and C-terminal part of C-CDNF (161-187) (2) (1 eq.) were dissolved in NCL buffer [1.6 M (6 M Guanidinium chloride, 0.2 M phosphate, 25 mM MPAA, 50 mM TCEP, pH 6.4)] and the mixture was stirred at 40 °C overnight. After reaction completion (confirmed by LCMS) 250 vol% Et₂O was added, and the mixture was shaken to extract excess of MPAA. The water layer was collected and purified by RP-HPLC and the product linear C-CDNF (127-–187) was collected. The linear product was then subjected for oxidation by dissolving in methanol/acetic acid (4:1 0.16 M) and Iodine (0.2 M in methanol) was added dropwise until the slight yellow color persisted. The reaction mixture was then stirred for another 30 min and the formation of cyclic product was confirmed by LC-MS. The solution was then concentrated under reduced pressure. The crude product was purified using RP-HPLC to yield the final pure C-CDNF (127–187). The specific conditions and data of the C-CDNF and its variants are detailed in Supplementary Table 1.

Synthesis of ¹³C/¹⁵N-Ala isotope labeled C-CDNF

For the synthesis of the ¹³C/¹⁵N-labeled isotope labeled C-CDNF the above-described method was adopted. Two peptide fragments were synthesized using same procedures for all amino acids except the labeled alanine residues. (A135, A145, A162, A175, A179, A180). These specified amino acids were incorporated manually using Fmoc-Ala(¹³C₃, ¹⁵N)-OH (2 eq., 0.4 M in DMF) HCTU (2 eq., 0.5 M in DMF), and DIEA (4 eq., 0.5 M in DMF) for 3 h at RT. More details of the isotopic C-CDNF can be found in Supplementary Table 1.

Synthesis of C-CDNF analogs

Synthesis of C-CDNF mutants (MT1, MT2 and MT3) were synthesized following the same protocols used for C-CDNF WT. All the conditions such as coupling, deprotection, cleavage, purification, NCL and the S-S bridge were the same. Details of the yield of all the mutants synthesized and their purity data is included in the Supplementary Table 1.

Isothermal titration calorimetry

ITC data was collected with MicroCal PEAQ-ITC (Malvern Panalytical, UK). GRP78-NBD and CDNF were dialyzed overnight against PBS at pH 7.4. Data was collected according to manufacturer’s recommendations. 300 μl of GRP78-NBD was used at a concentration of ~26 μM in the sample cell, while CDNF was injected using a concentration in the range of 300–500 μM. In addition, a reference injection into buffer was performed. 12 injections of 3 μl were made into the sample cell with a delay of 150 s between injections. Curves were analyzed using instrumental software and assuming a one-site binding model.

Spectral shift assays

A serial dilution of the ligands (non-fluorescent molecule; C-CDNF and C-CDNF mutants) was prepared in a way to match the final buffer conditions in the reaction mix (assay buffer: 1 x PBS pH 7.2, 0.05% Tween 20). The highest concentration of ligand was 200 μM and the lowest 6.10 nM. 5 μl of each ligand dilution sample were mixed with 5 μl of the target (GRP78-NBD with NT-650-NHS fluorescent molecule). The final reaction mixture, which was filled in premium capillaries, contained a respective amount of ligand (max. conc. 100 μM, min. conc. 3.05 nM) and a constant 10 nM fluorescent molecule tagged target molecule. The samples were analyzed on a Monolith X at 25 °C, with 100% LED power.

Figure 4B and Supplementary Fig. 5 show the fluorescence ratio signal (670 nm / 650 nm) for the repeat measurements and the raw data for each ligand concentration. In case the data points cannot be fitted to obtain a binding curve, its concluded to be ‘no interaction’ (Fig. 4B). If the data points allow for a high-quality curve-fit by using either the Hill equation or the K_D fit derived from the law of mass action according to the binding model, the affinity is calculated and stated as EC₅₀ or K_D values as indicated in Fig. 4B. The amplitude (shown in Fig. 4B) is the absolute difference between the fluorescence ratio signal (670 nm / 650 nm) of the “unbound” plateau at low ligand concentrations and the “bound” plateau at high ligand concentrations. The fluorescence ratio values for each ligand concentration are plotted in a semi-logarithmic manner against the ligand concentration to yield a dose-response curve (Fig. 4B).

NMR experiments

Commercial reagents were purchased from Medicago, Cambridge Isotope Laboratories, and Deutero and were used as received. NMR data were recorded on a 500 MHz Varian Inova spectrometer equipped with a 3 mm ¹H/¹³C/¹⁵N triple-probe. DSS (4,4-dimethyl-4-silapentane-1-sulfonic acid) was used as the chemical shift reference standard for ¹H and ¹³C chemical shifts, with the dimethyl signal set to δ 0 ppm in both dimensions. The spectra was processed using the software MestReNova (version 15.0.0-34764).

The 1D ¹H data were collected by applying a 45° pulse and using suppression of the water signal with the standard sequence “presat” in VnmrJ 4.2. The spectral width for the 1D ¹H experiments was 14.5 ppm, and the relaxation delay was 0.4 s. The acquisition time was 0.4 s, with a total of 64 scans collected into 64 k data points. A gradient version of the 2D ¹H-¹³C HSQC experiment (standard sequence in VnmrJ 4.2) optimized for the methyl fingerprint region was used. Data were recorded with spectral widths of 14.5 ppm in ¹H and 70 ppm in ¹³C with the ¹³C carrier at 8 ppm, a relaxation delay of 1.40 s, acquisition times of 0.02 s and 0.09 s in the ¹³C and ¹H dimensions, respectively, 128 number of increments in the ¹³C dimension, and 64 scans per increment. ¹⁵N NMR data were also acquired but those data were not used due to insufficient spectral quality during the titration (Supplementary Fig. 8).

NMR titration experiments were performed at 298 K (25 °C) as follows. A stock solution of the ¹³C₃/¹⁵N-Ala labeled C-CDNF was prepared by mixing the material (2.446 mg, 0.317 µmol) with PBS pH 7.2 (900 μL), D₂O (94 μL) and DSS (10 mM in D₂O, 6 μL). The solution was kept at 278 K (5 °C) prior to the titration. The prepared C-CDNF stock solution (56.6 μL) was diluted with additional PBS pH 7.2 and D₂O 9:1 (143.5 μL) and transferred to a 3 mm NMR tube. 1D ¹H and 2D ¹H-¹³C HSQC data were collected. Aliquots of tag-less GRP78-NBD in PBS pH 7.2 (9.65 mg/mL, 20 μL) was added to the NMR tube using a Hamilton syringe. 1D ¹H and 2D ¹H-¹³C HSQC data were collected for the following titration points: 0, 21 µM, 38 µM, and 65 µM of GRP78-NBD (Fig. 5).

Residue-specific assignment of the ¹³C-labeled methyl groups of the alanine residues was achieved based on the previously reported ¹H/¹³C/¹⁵N chemical shift assignments of the N-terminal domain of human CDNF⁴⁴. Although there are some deviations due to the different sample conditions including a lower pH of 6.0 as compared with this study, the chemical shift values for the assigned alanine methyl groups seem consistent with the reference data. An estimated interval of the dissociation constant for the binding of C-CDNF to GRP78-NBD was obtained from fitting the measured peak areas of A162 and A175 to the equations for bimolecular binding as further described in the Supplementary Table 3.

In vitro neuroprotection model

Rat primary mesencephalic neurons were prepared as described previously²⁸. On DIV6, the culture medium [Neurobasal (Invitrogen) supplemented with B27 (2%), L-glutamine (2 mM), 2% penicillin-streptomycin and 10 ng/ml of brain-derived neurotrophic factor (BDNF) and 1 ng/mL of glial cell line-derived neurotrophic factor (GDNF)] was replaced with fresh medium or with medium containing 4 µM 1-methyl-4-phenyl-pyrididium (MPP⁺; Sigma Aldrich), a mitochondrial toxin widely used to model PD-like dopamine neuron degeneration²⁷, for the following 48 h. In the treatment groups, the exposure of cell culture to MPP⁺ was done in the presence of test compounds (CDNF, C-CDNF or mutant proteins). Stock solutions of test compounds were prepared in PBS, diluted in culture medium, and then incubated with the neuronal culture starting from 4 h before MPP⁺ injury. The study was designed to include a control group without MPP⁺ and an MPP⁺ group without test compounds to each plate for controlling the inter-plate variability. Based on previous experience with standardized experimental conditions in this assay, 6 wells per condition was used as a technical replicate. Assay variability, as determined by %CV between replicate experiments, was 14.4% and 14.8% for control and MPP⁺ conditions, respectively (8 replicates, a total of 48 wells for each). For the positive control (C-CDNF), assay variability was 9.1% and 8.7% at concentrations 5 and 50 nM, respectively (5 replicates, a total of 30 wells for each; Source Data File).

48 h after the MPP⁺ injury, the culture medium was removed, and the cells were fixed in 4% paraformaldehyde in PBS, pH 7.3 for 20 min at RT. The cells were washed twice in PBS and then permeabilized and blocked by incubation in a solution of PBS containing 0.1% saponin and 1% FCS for 15 min at RT. The cultures were then immunostained with a mouse anti-tyrosine hydroxylase (TH; Sigma-Aldrich, Cat. No. T1299, Lot No. 122972, 1:10000) and a rabbit anti-α-synuclein (aSyn; Ozyme, Cat. No. 2642S, Lot No. 5, 1:200) antibodies in PBS containing 1% FCS and 0.1% saponin for 2 h at RT, followed by Alexa Fluor 488-conjugated goat anti-mouse IgG (1:800) and Alexa Fluor 568-conjugated goat anti-rabbit IgG (1:400) in PBS containing 1% FCS and 0.1% saponin for 1 h at RT. For markers of UPR activity, the cultures were double-stained with the mouse anti-TH antibody with either a rabbit anti-phospho-IRE1α antibody (Diagomics Oncology Expert, Cat. No. AF7150-200, Lot No. 32U2444, 1:500), a rabbit anti-ATF4 antibody (ThermoFisher, Cat. No. PAS27576, Lot Nos. ZA4195580C and YK4116700E, 1:200), or a rabbit anti-ATF6 antibody (Fisher Scientific, Cat. No. 12898242, Lot No. XJ3738203, 1:500) in a buffer containing 1% FCS and 0.1% saponin, followed by staining with corresponding fluorophore-conjugated anti-mouse and anti-rabbit secondary antibodies. Cell nuclei were counterstained with the fluorescent dye Hoechst (Sigma-Aldrich, 1:1000). For each test condition, 20 pictures (representative of the whole well area) were automatically acquired using ImageXpress® (Molecular Devices) at 10x magnification using the same acquisition parameters. From images, the analyses were automatically performed by MetaXpress® software (Molecular Devices) using the following readouts: total number of TH⁺ neurons, total length of neurite network of TH⁺ positive neurons (in μm), aSyn aggregation (overlapping TH and aSyn staining in μm²), phospho-IRE1α (overlapping TH and phospho-IRE1α staining), nuclear ATF4 (stained nuclei in TH⁺ cells) and nuclear ATF6 (stained nuclei in TH⁺ cells). Results are expressed as percent change vs untreated control condition. Statistical analysis of the different conditions was done by using GraphPad Prism software (version 8.0.2). One-way ANOVA followed by Dunnett’s test was used for group comparisons (with assumption of equal SDs). Brown-Forsythe test and Bartlett’s test were used to assess equality of group variances. Data normality distribution was assessed with Kolmogorov-Smirnov test.

The experiments were carried out in compliance with the current European Union regulations (Directive 2010/63/EU) in a facility that holds a valid accreditation issued by the French Ministry of Agriculture and were approved by the veterinary services of the Bouches-du-Rhone (license number B1301310).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

A.B. and H.J.H. conceptualized and planned the study. M.A.G., A.B., T.G. and D.S. planned and executed the SAXS and related experiments. M.V., S.M., J.A.E.M. and R.E.J. contributed to development of compounds and other tools used in the study. K.J., A.B. and J.E. planned and executed the NMR experiments. A.B., H.J.H., N.K. and M.A.G. analyzed and interpreted the data. A.B., M.A.G. and H.J.H. wrote the manuscript with inputs from all of the authors.

Peer review

Peer review information

Nature Communications thanks Brandon Harvey and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All relevant data associated with the published study are present in the paper or the Supplementary Information. Source data are provided with this paper. HPLC and Mass Spectrometry Data are provided in Supplementary Data 1. The SAXS data generated in this study have been deposited in the SASBDB database under accession code SASDNW9 (GRP78-NBD only; https://www.sasbdb.org/data/SASDNW9/), SASDQS6 (GRP78-NBD:C-CDNF complex; https://www.sasbdb.org/data/SASDQS6/), SASDNX9 (GRP78-NBD:CDNF; https://www.sasbdb.org/data/SASDNX9/). The processed SAXS data are available at SASBDB. The NMR data used in this study are available in the BMRB database under accession code 52561. Previously published accession codes used in this study: 3IUC.pdb (crystal structure of the monomer of GRP78-NBD) [10.2210/pdb3IUC/pdb]. 4BIT.pdb (solution NMR structure of CDNF) [10.2210/pdb4BIT/pdb]. 6HA7.pdb (crystal structure of GRP78-NBD in complex with the CDNF homolog MANF) [10.2210/pdb6HA7/pdb]. Source data are provided with this paper.

Competing interests

N.K., A.B. and H.J.H. are employees and share or option right holders of Herantis Pharma Plc, who sponsored the study. M.V., K.J., S.M., J.E., and R.E.J. are employees of Red Glead Discovery AB. T.G. and D.S. are employees of BioSAXS GmbH. N.K., A.B. and H.J.H. are inventors in two international patent applications that are related to compounds that were developed partially based on the structural data shown in this paper. The applications are pending and have the application numbers: W.O. 2021/123047A1 and WO 2021/123050A1. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-024-52478-0.