Protective anti-prion antibodies in human immunoglobulin repertoires

Prion immunotherapy may hold great potential, but antibodies against certain PrP epitopes can be neurotoxic. Here we identified >6000 PrP-binding antibodies in a synthetic human Fab phage display library, 49 of which we characterized in detail. Antibodies directed against the flexible tail of PrP conferred neuroprotection against infectious prions. We then mined published repertoires of circulating B cells from healthy humans and found antibodies similar to the protective phage-derived antibodies. When expressed recombinantly, these antibodies exhibited anti-PrP reactivity. Furthermore, we surveyed 48’718 samples from 37’894 hospital patients for the presence of anti-PrP IgGs, and found 21 high-titer individuals. The clinical files of these individuals did not reveal any enrichment of specific pathologies, suggesting that anti-PrP autoimmunity is innocuous. The existence of protective anti-prion antibodies in unbiased human immunological repertoires, combined with the reported lack of such antibodies in carriers of disease-associated PRNP mutations, suggests a link to the low incidence of spontaneous prion diseases in human populations.


Introduction
Many neurodegenerative syndromes, including prion diseases, Alzheimer's disease and Parkinson's disease, go along with the accumulation of misfolded and aggregated proteins in the central nervous system. Antibodies against such proteins may be beneficial 1 , e.g. by opsonizing pathological aggregates and mediating their degradation by phagocytic cells 2,3 .
While the clinical effectiveness of antibody-based therapies against neurodegenerative diseases is still being debated 4 , there is ample evidence that both active immunization and passive antibody transfer can effectively clear pathological aggregates in preclinical animal models and, to some extent, in affected humans.
According to the protein-only hypothesis, the prion is an infectious particle consisting of PrP Sc , an aggregated and proteinase K (PK) resistant isoform, of the cellular prion protein PrP C 5 .
PrP C consists of a C-terminal globular domain (GD) and an N-terminal flexible tail (FT) which includes the octapeptide repeat (OR) region, two cationic charge clusters (CC1 and CC2) and a hydrophobic core (HC) 6 . The CC1 domain of PrP C participates in Schwann cell maintenance by activating the G protein-coupled receptor Adgrg6 7,8 . While PrP C deficient mice are only mildly affected, PrP Sc necessitates PrP C for its propagation 9 and prion toxicity is transduced by PrP C onto target cells 10,11 . Therefore, suppression of PrP C by means of anti-PrP C antibodies represents a rational strategy against prion diseases.
Here we panned a synthetic human antibody phage display library to explore the presence of PrP-binding antibody fragments (Fabs) 12 . To identify rare antibodies to poorly antigenic epitopes that may be overlooked by conventional screening technologies, we performed "nextgeneration" sequencing (NGS) of panning outputs after phages selections 13 . Several anti-PrP binders were identified and found to antagonize prion toxicity. What is more, mining of published human antibody repertoires identified sequences similar to an anti-prion phagederived Fab which, when expressed, acted as functional PrP binders. Lastly, the interrogation of a large unselected hospital cohort (n = 37'894) highlighted individuals with high-titer anti-PrP autoreactivity whose clinical presentation was heterogeneous, yet unrelated to known features of prion diseases. Therefore, anti-prion immunity can exist in human communities and is seemingly innocuous.

Phage display selection strategy for anti-PrP Fabs
We used three rounds of phage display to screen two synthetic human Fab phage display libraries (Extended data Fig. 1a) with short (8-10 aa) and long (12-20 aa) heavy-chain complementarity-determining regions 3 (HCDR3). These libraries were constructed to mimic human antibody repertoires by combining frameworks from human germline sequences with diversified HCDR3 whose design approximated the natural gene sequences in human repertoires as compiled in the IMGT database 14 . The first and second biopanning rounds were performed against full-length recombinant mouse PrP (recPrP23-231) to enrich for Fabs covering a large variety of PrP epitopes. To further select Fabs recognizing specific PrP epitopes, a third panning round was conducted against several different antigens in parallel, including recPrP23-231, recPrP23-110 spanning the FT, the GD (recPrP90-231 and recPrP121-231) and synthetic peptides representing CC123-50 spanning the CC1, N-OR39-66 and F-OR51-91, containing the OR, and CC2-HC92-120, spanning CC2 and the HC.
Only few existing antibodies bind to the natively unstructured CC123-50 15 . To optimize our chances to identify CC123-50 binders, and to avoid misfolding artefacts caused by nonspecific plate adsorption, the selection against the biotinylated CC123-50 peptide was conducted in solid phase and by liquid phase panning followed by neutravidin (Neu)-mediated capture. Liquidphase panning was also performed for other biotinylated peptides (N-OR39-66, F-OR51-91 and CC2-HC92-120). Furthermore, we performed panning rounds using a matrix of stringent washing conditions (Extended data Table 1) to also include an affinity read-out to the analysis of the NGS screening.

Next Generation Sequencing (NGS) of anti-PrP Fabs
The HCDR3 domains can contribute crucially to antigen binding 16 . Sequencing of the outputs of the third panning rounds yielded 4,847 and 11,948 unique HCDR3 sequences in 100'000 analyzed sequences for the short and long HCDR3 Fab libraries, respectively. We excluded all HCDR3 that had any counts in the negative-control outputs (Neu and BSA panning) and retained only clones with ≥ 1 read in the recPrP23-231 output. These constraints reduced the unique HCDR3 sequences to 1,173 and 4,832 anti-PrP Fabs, respectively. We then compared the read counts of each HCDR3 between panning to recPrP23-231 and the different PrP domains. For each HCDR3, we considered the enrichment of NGS counts in a panning output as reflecting the binding to the respective PrP peptide/fragment used in the panning. As an example of the stringent sorting criteria for epitope binding profile determination (Extended   data Table 2), all HCDR3 having counts > 0 in CC123-50 and in recPrP23-110 outputs, and count = 0 in all other panning outputs, were classified as specific PrP binders in the CC123-50 region.
HCDR3 sequences were clustered based on their NGS-binding profile and found to represent a highly diverse collection of anti-PrP Fabs (Fig. 1a, b). Predominant clones binding to the high antigenic epitopes, defined as those with ≥ 20 NGS counts in recPrP23-231 panning, were mostly directed against the CC2-HC92-120 in both HCDR3 libraries (Fig. 1c). Rare clones against less antigenic epitopes, i.e. displaying only one count in recPrP23-231 panning, were predominantly showing an NGS-binding profile to the CC123-50 and the GD domains (Fig. 1c).
We retrieved clones of interest, as identified by the HCDR3 read profile in NGS, by overlapping PCR from the third-round polyclonal phagemid DNA (Extended data Fig. 1b-e). In one instance we designed primers to an HCDR3 sequence with an NGS-binding profile to the CC2-HC92-120 epitope (NGS read enrichment in the CC2-HC92-120 panning as compared to reads in panning to other PrP domains, Extended data Fig. 1b) and retrieved entire Fab sequences by PCR (Extended data Fig. 1c). The retrieved Fab, designated as FabRTV, was cloned into an E. coli expression vector, Sanger-sequenced and purified by immobilized metal ion affinity chromatography (IMAC) (Extended data Fig. S1d). Enzyme-linked immunosorbent assay (ELISA) confirmed the NGS-binding profile of FabRTV to the CC2-HC92-120 domain (Extended data Figs. 1b, e).

Identification of anti-PrP Fabs by ELISA screening
As a complementary approach, we screened 4416 clones (randomly selected by an automated colony picker) by ELISA. We then selected 312 hits reactive to recPrP23-231, to recPrP fragments, or to synthetic biotinylated peptides (>10 or >5-fold over background, while displaying a Neu signal <5-fold over background). From those, eighty confirmed anti-PrP Fabs were Sanger-sequenced, produced in E. coli and epitope-mapped by ELISA (Extended data  Table 3). The abundance of CC2-HC92-120 targeting Fabs is in agreement with the NGS analysis showing that CC2-HC92120 binders are the predominant clones. For 35 of these Fabs, the ELISA binding results confirmed the epitope binding profile determined by NGS analysis.

Kinetic measurements by surface plasmon resonance (SPR) and affinity maturation
The KDs (determined by SPR as koff/kon ratio) of all tested Fabs to recPrP23-231 were in the range of 10-100 nM (Table 1) with fast dissociation rate constants (koff > 10 -3 [s -1 ]).
To optimize the binding properties of Fab3 and Fab71, to CC123-50 and OR51-91 respectively, we used affinity-maturation libraries in which the parental HCDR2 and LCDR3 loops were replaced with pre-built highly diversified cassettes. In addition, phage display selections were repeated with more stringent conditions than in the original selection. To ensure retention of specificity to the respective epitopes, panning against recPrP23-231 was alternated with panning to the CC123-50 fragment for Fab3 and to the OR51-91 fragment for Fab71. Again, selected Fabs were screened by NGS and ELISA. After high-throughput off-rate and on-rate ELISA screening, 19 and 5 affinity-matured versions of Fab3 (Fab81-Fab99) and Fab71 (Fab100-Fab104) were identified, respectively. These clones were also prevalent in the NGS dataset.
With this strategy, the EC50s of the affinity-matured Fabs were improved by 2-3 logs over the parental Fabs (Fig. 2a and Extended data Table 4).

Epitope confirmation and mapping
Next, we confirmed the binding behavior of the different Fabs to biotinylated FT-PrP peptides CC123-50, N-OR39-66, F-OR51-91 and CC2-HC92-120 by SPR (Extended data Fig. 3). For all tested Fabs, the specificity of the targeted epitope matched the ELISA epitope profiling. Being unstructured, the FT displays many linear epitopes 17 . Therefore, we additionally mapped the epitopes of the anti-FT Fabs by competition ELISA using overlapping dodecameric PrP peptides (each shifted by 2 residues) spanning residues 23-120. The binding of Fab3 and of its affinity-matured version Fab83 to immobilized recPrP23-231 was blocked by peptides P1 and Fab44 recognized the sequence WGQPHGG within the OR (Extended data Fig. 4). For Fab7, Fab10, Fab41 and Fab52, we could not identify any PrP peptide that abrogated the ELISA signal to recPrP23-231. This may point to the presence of conformational/discontinuous epitopes.
In addition, Fab83, Fab100, Fab53 and Fab74 were also able to detect wild-type PrP C (wtPrP C ) in brain homogenates (BH) from wild-type (C57BL6/J) and tga20 mice 18 overexpressing wtPrP C , but not in the brains of mice expressing PrP deletion mutants lacking the respective epitopes (Extended data Fig. 5b).
The CC123-50 binder Fab83 did not detect PrPΔ25-50 19 and PrPΔ32-93 20 on Western blots, nor did it detect cell expressed PrPΔ23-27, PrPΔ23-31 and PrP mutated in the polybasic stretch from KKRPK to KPRKK by ELISA (Extended data Fig. 5c). Lysine-to-alanine substitutions within the KKRPK sequences reduced the binding of Fab83. Hence the Fab83 epitope comprises both lysine K24 and proline P26 in the N-terminal KKRPK stretch (PrP23-27) and additional residues in the PrP32-36 segment. Alternatively, KKRPK could have become buried in the truncated version PrPΔ32-93.
Consistent with the peptide epitope mapping, Fab100 did not react with the OR-deleted PrPΔ32-90 and PrPΔ51-90 19 . In addition, Fab53 did not detect PrPΔ94-110 7 , whereas the anti-GD Fab74 recognized all FT-modified versions of PrP. Most of the tested Fabs also cross reacted with human recPrP23-230 (Extended data Fig. 5a).
We then performed immunoprecipitations of wtPrP C from BH of wt mice and from mice expressing different PrP deletion mutants, with Fab83 (CC123-50) and Fab71 (OR51-91) (Extended data Fig. 5d,e). Fab83 and Fab71 immunoprecipitated wtPrP C but not PrPΔ25-50 and PrPΔOR with deletion of the Fab binding sites as predicted from the ELISA. wtPrP C was eluted by the P1 peptide containing the Fab83 binding site within the CC123-50 but not by the P15 peptide within the OR51-91 (Extended data Fig. 5d). Similarly, Fab71 immunoprecipitated wtPrP C was eluted from the Fab71-beads complex by the epitope-mimicking peptide P17 in the OR51-91, but not by the P3 peptide within the CC123-50 (Extended data Fig. 5e).
We then tested Fab100 for immunoprecipitation of PrP Sc from prion infected brain homogenates (Fig. 2f). Fab100 immunoprecipitated PrP C and PrP Sc from NBH and prioninfected brains, respectively. Elution under native conditions was achieved using peptide P17 (within OR51-91) but not P2 (within CC123-50), confirming the specificity of Fab100 for the OR of both PrP C and PrP Sc . Proteinase-K (PK) digestion assays confirmed the presence of PrP Sc in the eluted fractions from prion-infected BH after immunoprecipitation by Fab100 (Fig. 2g).

Validation by flow cytometry and immunostaining
We also assessed the binding of a panel of Fabs to cell-surface wtPrP C by flow cytometry using the murine neuroblastoma cell line CAD5. For each Fab, we compared the mean fluorescence intensity (MFI) between CAD5 Prnp +/+ and Prnp -/cells (Extended data Fig. 6a-b). All CC123-50 binders, the majority of OR51-91 binders and CC2-HC92-120 binders discriminated between Prnp +/+ and Prnp -/-CAD5 cells, with Fab71 being the best performer. Of the GD binders, only Fab74 showed a differential signal, suggesting that the GD of membrane-bound wtPrP C is less accessible to antibodies than the FT. Binders were tested by immunofluorescence on paraffinembedded brain sections of C57BL6/J, Prnp ZH3/ZH3 and tga20 mice. Fab3, Fab71 and Fab74 detected murine wtPrP C (Extended data Fig. 6c) with high specificity.

Activity of Fabs in models of prion disease
Antibodies that bind to the FT have been demonstrated to be neuroprotective, whereas those directed against certain epitopes of the GD are invariably toxic 11,21 . We therefore asked whether Fabs against CC123-50 (Fab3 and Fab83), the OR51-91 (Fab8, Fab44, Fab71 and Fab100) and the GD (Fab25, Fab1 and Fab29) may counteract neurotoxicity in prion-infected cerebellar slices (COCS) 11,22 . Tga20 COCS were exposed to Rocky Mountain Laboratory prions (passage #6, RML6) or to NBH and cultured in the presence or absence of the different Fabs (550 nM). At 45 days post-infection (dpi), prion-infected COCS showed conspicuous neurodegeneration ( Fig. 3a-c). The GD binder Fab25 showed intrinsic neurotoxicity as indicated by loss of NeuN immunoreactivity in NBH-treated COCS. All tested OR51-91 binders, but none of the GD binders prevented prion-induced neurotoxicity. Fab3 (binding to CC123-50) did not prevent prion neurotoxicity, whereas its affinity matured derivative Fab83 was effective (Figure 3a-c).
We then investigated the effects of the Fabs on prion replication and PrP Sc accumulation (Fig.   3d). Treatment with the anti-OR51-91 Fab100, but not by anti-CC123-50 Fab83, reduced PrP Sc in prion-infected COCS, although both were neuroprotective. In slices treated with the GD Fab25, PrP Sc levels were not significantly different from untreated prion-inoculated slices.
We also assessed the ability of the Fabs to arrest prion replication and/or PrP Sc accumulation in prion susceptible cells. CAD5 Prnp +/+ and CAD5 Prnp -/cells were exposed to prions or NBH and treated with 10 µg/ml of either CC123-50 binder Fab3, OR51-91 binder Fab71 or the GD targeting Fab29. Fabs were added to the medium 1 hour after infection and after every splitting. At 14 dpi, PrP Sc content was determined using a homogenous-phase PrP Sc time resolved (TR) FRET quantification immunoassay after proteolytic PrP C removal. PrP Sc was seen by TR-FRET in prion infected CAD5 Prnp +/+ but not in CAD5 Prnp -/cells indicating that the initial prion inoculum had been diluted below detectability. Fab83 did not reduce PrP Sc level in CAD5 Prnp +/+ cells despite being protective in COCS (Extended data Fig. 7). Fab71 and Fab100, but not Fab3 and Fab29, substantially lowered PrP Sc in prion infected CAD5 Prnp +/+ cells ( Fig. 3e and Extended data Fig. 7). Hence OR51-91 binders exerted their neuroprotective activity by reducing prion accumulation.

Identification of anti-PrP antibodies from human antibody repertoires similar to the phagederived Fabs
The Fabs were derived from a synthetic library mimicking the human antibody repertoire, raising the question whether analogous antibodies might be present in bona fide human repertoires. The heavy and light-chain frameworks of Fab71 correspond to human germlines VH3-30 and Vκ3-15, respectively, whereas its HCDR3 originates from the library randomization strategy. We therefore searched for human antibodies harboring HCDR3 amino acid sequences similar to that of Fab71. We examined large-scale repertoire datasets with billions of sequences that were generated recently by sequencing heavy-chain (VH; here referred to as DW 23 and BB 24 datasets) and natively paired heavy/light-chain variable regions (VH:VL; DK dataset) 25 from circulating naïve and memory B cells of healthy donors. While no exact match to Fab71 HCDR3 was found, twelve out of fourteen analyzed donors displayed 629 HCDR3 sequences that differed from Fab71 HCDR3 by ≤ 3 residues (Fig. 4a). Among them, 74 HCDR3 sequences occurred in ≥ 2 subjects, four of which differed from Fab71 HCDR3 by only one residue (Fig. 4a).
We selected eight heavy chain variable regions similar to Fab71. In one of these, HCDR3 deviated from Fab71 in just two residues and matched the Fab71 VH3-30 segment. Three further HCDR3 differed from Fab71 HCDR3 by only one residue and two of those HCDR3 appeared in antibodies using different V genes (VH3-21 and VH1-46) from three subjects (Fig.   4b). We expressed these heavy chains variable regions in bacteria along with the Fab71 light chain to purify Fabs and test their reactivity to human recPrP23-230. In the case of DK_VH3-07, where VL sequence information was available from the VH:VL paired sequencing dataset, we used VL_2-14. All selected naturally occurring Fab71-like antibodies, except the DK_VH3-07, recognized human recPrP23-230 by ELISA specifically (Fig. 4c). BB_VH3-30, DW_VH1-46 and DW_VH3-74 Fabs, displayed the highest apparent affinities (EC50 = 1.21 µM, EC50 = 1.07 µM and EC50 = 0.98 µM respectively; for DK_VH3-07 Fab, which did not bind to human recPrP23-230, calculated EC50 = 62.8 µM). While these affinities are low, they may be greatly raised by dimeric cooperativity in bivalent antibodies.

Identification of natural anti-PrP autoantibodies in an unselected hospital cohort
To assess the validity of the above findings, we interrogated a large cohort of human individuals for naturally occurring antibodies against PrP ( Fig. 5a and Extended data Fig. 8).
We performed an automated microELISA to screen 48'718 plasma samples from 37'894 individuals admitted to almost all clinical departments of the University Hospital Zurich ( Fig.   5a-b and Extended data Fig. 8a) for binding to human recPrP23-230. We applied stringent criteria: only plasma samples displaying log(EC50) ≥ -2 and logistic-regression fitting error <20% were considered as hits. In a primary high-throughput screen (HTS), 27 individuals (9 females and 18 males) were found to harbor IgG antibodies reacting to human recPrP23-230 ( Fig. 5c-d). A validation screen confirmed PrP reactivity in the plasma samples of 21 individuals, indicative of a 0.06% prevalence of autoantibody carriers (Fig. 5e). The clinical presentation of individuals with such autoantibodies was heterogeneous (Extended data Table   5), with no statistically significant enrichments found in disease codes (International Classification of Disease, ICD-10), medication reports, age or sex (Fig.5d, Extended data Fig.  8b,c and Extended data Table 6). None of the patients with anti-PrP reactivity showed signs of prion-related pathology. Repeated longitudinal sampling indicated that 3 individuals sustained their high anti-PrP titres over a time span of several months up to more than a year ( Fig. 5g), suggesting that these anti-PrP-autoantibodies are stable over time.
We then assessed the clonality of the anti-PrP-autoantibodies in a subset of PrP-reactive samples. First, we compared by ELISA the λ/κ light-chain ratio in total immunoglobulins vs.
anti-PrP autoantibodies. We found that antibodies binding human recPrP23-230 have preferential contribution of λ over κ light chains (Fig. 5f). We then purified immunoglobulins from selected plasma samples, confirmed their specific reactivity to human recPrP23-230, and assayed them for differential binding to recPrP23-110 (FT) versus recPrP121-230 (GD). While some patients showed immunoreactivity against the GD, antibodies in other patients targeted both the FT and the GD, suggesting a polyclonal antibody response (Fig. 5h). These findings corroborate the evidence for the existence of naturally occurring antibodies against the prion protein in humans.

Discussion
Anti-PrP antibodies are effective in cells and mice infected with prions 26,27 , and may represent a plausible therapeutic strategy 2,28 . However, although certain anti-PrP antibodies afford neuroprotection to prion-infected mice, others cause extensive neuronal loss 29 and several antibodies to the GD on PrP result in acute on-target toxicity 11,21 . Also, antibodies to the OR region of PrP prevent neurotoxicity triggered by GD-binding antibodies and by prions in organotypic slices 11,22,30 . Thus, the biological effect of anti-PrP antibodies crucially depends on the targeted PrP epitope. Hence, this study aimed to produce a high-resolution map of neuroprotective epitopes, with the ultimate goal of identifying effective immunotherapeutics.
The production of anti-PrP antibodies in wt animals is hindered by self-tolerance 31 . Therefore, anti-PrP monoclonal antibodies have been developed mainly in PrP-deficient mice 17,32 or by phage display 33 and target mostly immunodominant PrP epitopes within the central region and GD of PrP 34 .
Here we screened a synthetic human antibody library by phage display which differs from previous approaches 35,36 in several aspects. Firstly, we expanded the number of PrP antigens, instead of using a single PrP fragment for panning, with the intent to discover antibodies to all regions of PrP. Secondly, we deep-sequenced the panning outputs -a strategy that optimizes the detection of extremely rare antibody clones. Finally, we opted to bacterially express antibodies as Fabs which are typically more stable and less susceptible to dimerization than scFv 37 . This enabled us to generate anti-PrP Fabs with highly diverse specificities. In addition, this strategy yielded, besides the 49 Fabs identified by ELISA screening, hundreds of additional rare Fab hits against less antigenic epitopes of PrP.
Having produced a broadly diversified panel of anti-PrP Fabs, we assessed the correlation between their epitope and their biological activity. All OR51-91-binding Fabs prevented neurodegeneration in prion-infected COCS. These results are consistent with previous findings 11,22,30 pointing to the OR of PrP as the effector arm of neurotoxicity. In COCS and in prion-infected cells, neuroprotection by OR51-91 binders was associated with reduced levels of PK-resistant PrP Sc . This effect has not been described for POM2 that blocked the toxic cascade elicited by prions, downstream of its replication. Other OR-targeting antibodies 38 were described to block PrP C internalization, thus reducing the rate of intracellular conversion to PrP Sc . Alternatively, the engagement of the OR by Fab71 and Fab100 may prevent the interaction between PrP C and PrP Sc , as suggested for prion-clearing anti-GD antibodies in susceptible cell lines 26,27 .
CC123-50-targeting antibodies are rare and, to our knowledge, were never assessed for neuroprotection against prions. However, it is well-established that the CC1 is required for the neurotoxicity of PrP mutants with deletion in the central domain 39 . Transgenic mice expressing PrPΔ23-31 or PrPΔ23-88 not only exhibit prolonged survival after prion inoculation but also accumulate less PrP Sc in their brains 40,41 . Conversely, Fab83 did not reduce PrP Sc levels in CAD5 cells and in prion infected COCS, yet it was neuroprotective. Furthermore, alanine substitutions of the positively charged amino acids within PrP 23-KKRPK-27 did not prevent PrP Sc formation in prion-infected neuroblastoma cells ScN2a 42 . We conclude that the antibodymediated neutralization of the CC1 is neuroprotective by other means than arresting PrP Sc generation. We speculate that the CC1 blockade prevents the interaction of PrP with downstream effectors of neurotoxicity, such as the Group-I metabotropic glutamate receptors 43 , which in turn trigger neurotoxicity.
The presence of anti-PrP antibodies in a human repertoire is surprising. Prions do not elicit antibody responses 44 , most likely because of the negative selection of B cells that autoreact to PrP C whose primary amino acid sequence is identical with PrP Sc 31 . Immunization of wt mice with recombinant PrP did not result in antibodies to N-terminal epitopes, and most antibodies recognized recombinant PrP in ELISA but not native PrP C on cell membranes. We found many Fabs binding within the FT in the synthetic human antibody-phage display library, most of which recognized native PrP on the cell surface. Unexpectedly, sequence comparisons identified PrP-reactive antibodies in published databases of naïve human repertoires from circulating B cells. Finally, we found high-titer PrP autoantibodies in the plasma of unselected hospitalized patients. Certain antibodies to PrP can mimic prion neurotoxicity 11,22 . At the time of the analysis, none of these subjects had received a diagnosis of prion disease. The clinical presentation of the patients with high-titer PrP autoantibodies was heterogeneous and, although signs of dementia were reported for two of the anti-PrP antibody carriers, we did not find any statistically significant correlation with neurological or any other disease when comparing subjects with and without anti-PrP autoantibodies. Similarly, PrP autoantibody titers present in a subset of PRNP mutation carriers neither correlated with the PRNP mutation status nor with the onset of clinical prion disease.
The frequency of high-titer anti-PrP antibody carriers (0.06%) is much lower than the occurrence of Fab71-like HCDR3 sequences in published human repertoires. This discrepancy could mean that most anti-PrP specificities exist in a dormant state, or are expressed as B-cell receptors, but do not produce circulating antibodies. It will be interesting to discover the triggers that may ignite antibody production and, possibly, afford protection against prions.
The evidence presented here indicates that the immunological repertoires of unselected humans can contain antibodies against PrP C . Such antibodies might protect against prions.
The rarity of individuals with high-titer anti-PrP immunity is unsurprising since PrP C is highly expressed in many immune cells including the developing thymus, which in mice results in an almost-insurmountable central tolerance 31 .
We suspect that it is not PrP C that induces adaptive immune responses against prions, but nascent PrP Sc instead. Accordingly, clinically silent prion generation may occasionally occur in healthy individuals 46 . PrP Sc aggregates arising de novo may result in exposure of neoepitopes and/or epitopes occluded in cell-borne PrP C . The resulting immune response may clear any nascent prions, akin to the immune surveillance against neoplastic cells. Progressive senescence of adaptive immunity 47 , which is well-documented for a variety of infectious diseases, may explain why both sporadic and familial prion diseases flare up mostly in late life.
In conclusion, the generation of antibodies to the whole PrP epitope space provides new tools to understand the mechanism of neurodegeneration conveyed by prions. The presence of prion protein binders in human antibody repertoires and of anti-PrP reactivity in human plasma points to a potential source of immunotherapeutics against prion diseases.
Recombinant mouse and human PrP proteins were produced as described 48 .

Construction of the synthetic human Fab phage library
A synthetic human Fab phagemid library (Novartis Institutes for BioMedical Research) was used for the phage display selections. A gene fragment encoding the germlines frameworks combinations IGHV3-23 and IGKV1-39, IGHV3-23 and IGLV3-9, IGHV3-30 and IGKV3-15, IGHV3-15 and IGLV1-47, were synthesized by Invitrogen's GeneArt service in Fab format and cloned into a phagemid vector serving as the base templates. These human germlines were used as they display favorable frameworks combinations for a phage display library 49 . The phagemid vector consists of Ampicillin resistance, ColE1 origin, M13 origin and a bi-cistronic expression cassette under a lac promotor with OmpA -light chain followed by PhoA-heavy chain -Flag -6xHis -Amber stoptruncated pIII (amino acids 231 -406).
Only HCDR3 was diversified and primers were designed to incorporate up to 11 amino acids at defined ratios mimicking their natural occurrences: aspartic acid, glutamic acid, arginine, histidine, serine, glycine, alanine, proline, valine, tyrosine, tryptophane. Leucine and phenylalanine were also allowed at a certain position of HCDR3. Certain residues were omitted on purpose to remove potential post translational modification sites. Randomized primer synthesis was performed using the Trinucleotide technology (ELLA Biotech) in order to exclude stop codons, methionine, cysteine and asparagine.
Lengths between 8, 10, 12, 14, 16 and 20 amino acids were allowed, in which the last two amino acids were kept constant with the sequence Asp-Tyr for length 8 to 16 and Asp-Val for length 20. The design of the final two HCDR3 amino acids reflects human VDJ recombination.
Short HCDR3s more often use J-fragment IGHJ4 with "DY" at the end of HCDR3, while longer HCDR3s (here 20 aa) more often use IGHJ6 with "DV" at the end of HCDR3.
Library inserts were generated by PCR using Phusion High Fidelity DNA polymerase (NEB Biolabs). The resulting HCDR3 library inserts were ligated into the base templates, transformed into E.coli TG1F+ DUO (Lucigen) with a minimal library size of 3E+08 transformants per HCDR3 length and phages were produced using VCSM13 helper phage (Agilent Technologies) using standard protocols.

Phage display for isolation of PrP binders
Depending on the HCDR3 length, the library was divided into two sub-pools for panning: short

Affinity maturation of selected Fab3 and Fab71
Fab3 and Fab71 were further engineered to improve their affinity by using affinity maturation cassette libraries (Novartis) with either diversification in the HCDR2 or in the LCDR3. The HCDR2 and the LCDR3 sequences repertoires were diversified according to naturally occurring repertoire of rearranged human CDR sequences.
For LCDR3 libraries, primers were designed to incorporate up to 11 amino acids at defined ratios mimicking their natural occurrences: aspartic acid, glutamic acid, arginine, histidine, threonine, serine, glycine, alanine, leucine, valine, tyrosine. Glutamine, proline and tryptophan were also allowed at certain positions of LCDR3. LCDR3 lengths of 9 and 10 amino acids were allowed for IGKV1-39 and IGKV3-15, in which the last threonine was kept constant.
For HCDR2 libraries, primers were designed to incorporate up to 10 amino acids at defined ratios mimicking their natural occurrences: aspartic acid, glutamic acid, arginine, histidine, threonine, serine, glycine, alanine, valine, tyrosine. Isoleucine, and tryptophan were also

Sample preparation for NGS
Polyclonal DNA minipreps isolated from the third panning output pools were used as PCR template to amplify the HCDR3 of the selected Fabs and add the adapters required for sequencing on Illumina sequencer MiSeq. The PCR protocol has been described 50 .
The PCR product was purified and DNA concentration was measured using the Qubit DNA High sensitivity kit (Invitrogen). Samples were analyzed on a MiSeq using MiSeq reagent kit MiSeq v2 Reagent kit 300 cycles PE.

NGS data analysis
The data analysis of the NGS FastQ output files was performed as described 50 . For each panning output, 100'000 sequences were analyzed using the fixed flanking sequences on the boundary of variable region as template to locate and segment out the HCDR3 sequence.
~40'000 to 70'000 HCDR3 sequences were identified depending on the panning output pools, and included into frequency reports in CSV format.
For determination of clones to high immunogenic PrP epitopes, we selected HCDR3 displaying ≥ 20 NGS counts in recPrP23-231 panning in 100'000 analyzed sequences. For rare clones against less immunogenic PrP epitopes, HCDR3 were identified according to the following criteria: NGS count in recPrP23-231 panning = 1 and, to avoid selecting for sequences resulting from PCR or sequencing errors, sum of the NGS counts across all the panning outputs ≥ 10.  Competition ELISA for epitope mapping

Rescue of clones identified in NGS
The approach described by Polymenidou et at. 17  For Fabs that were identified by mining of NGS datasets of human antibody repertoires, human recPrP23-230 at 87 nM in PBS was used for coating and serially diluted Fabs were tested starting from 3 µM in PBS-T.

Measurement of binding kinetics by surface plasmon resonance (SPR)
Binding kinetic of the purified Fabs to full length PrP was monitored at 25 °C using ProteOn

Establishment of CAD5 Prnp -/cells by CRISPR
The CAD5 cell line derived from Cath.a-differentiated (CAD) cells was reported to be responsive to several prion strains 51  Eluate was finally collected and supplemented with loading buffer (NuPAGE, Invitrogen) for western blot analysis using the mouse monoclonal antibody POM1 and the rabbit polyclonal antibody XN (produced in house) for PrP detection.

Immunoblot analysis
For epitope confirmation of the Fabs by western blot, brain from transgenic mice expressing different PrP deletion mutants was homogenized in 10 volumes of lysis buffer (50 mM Tris-HCl pH 8, 0.5% Na deoxycholate, and 0.5% Igepal, protease inhibitors (complete Mini, Roche) using TissueLyser LT for 5 min for 2 cycles. After centrifugation at 1000 x g for 5 min at 4°C to remove debris, protein concentration in the post nuclear supernatant was measured by BCA. Samples were adjusted to 20 µg protein in 20 µl and digested with 5 µg/ml proteinase-K (PK) in digestion buffer (0.5% wt/vol sodium deoxycholate and 0.5% vol/vol Nonidet P-40 in PBS) for 30 min at 37°C. PK digestion was stopped by adding loading buffer (NuPAGE, Invitrogen) and boiling samples at 95°C for 5 min. Proteins were separated on a 12% Bis-Tris polyacrylamide gel and blotted onto a PVDF membrane by using the iblot apparatus (Bio-rad).

Antibody treatment in cultured organotypic cerebellar slices
Cultured organotypic cerebellar slices were prepared from 9-12 day old tga20 pups as previously described 52 . For prion experiments, COCS were infected as free-floating sections with 100 µg per 10 slices of RML6 (Rocky Mountain Laboratory strain mouse-adapted scrapie prions at 6 passage) brain homogenate from terminally sick prion-infected mice. As control, non-infectious brain homogenate (NBH) from CD1-inoculated mice was used. After incubation with brain homogenates diluted in physiological Grey's balanced salt solution for 1 h at 4°C, the slices were washed and 5-8 sections were seeded on a 6-well PTFE membrane insert.
Treatment with Fab (550 nM) was started 1 day after plating and supplied at every medium exchange. At 45 days in culture, slices were fixed and processed for immunocytochemistry.

Prion-infected CAD5 cells
CAD5 Prnp +/+ and CAD5 Prnp -/cells were cultured in phenol red free OPTI-MEM supplemented with 10% FBS, Glutamax, penicillin G and streptomycin at 37°C in 5% CO2/95% air. CAD5 Prnp +/+ and CAD5 Prnp -/cells (5 × 10 4 in 2 ml of medium) were seeded into 6-well plates (Corning Costar) and cultured for 1-2 days before exposure to 500 µg/ml of prioninfected mouse brain homogenate. Non-infectious brain homogenate (NBH) from CD1 mice at the same dilution was used as control. Treatment with Fabs at 10 µg/ml (200 nM) was initiated 2h after infection and repeated at every split by spiking into the culture medium. Three biological replicates were prepared for each condition. The inoculum was removed after 3 days and the cells were split 1:5 every 3-4 days. After 4 splits (14 days in vitro, DIV) the cells were assayed for PrP Sc by the TR-FRET assay as described below. The collected fluorescence data were corrected by both background and spectral overlap between Eu and APC channel. Net FRET calculations and blank subtractions were performed as previously described 53 .

Identification and cloning of Fabs from human antibody repertoire datasets
NGS datasets of antibody repertoires from published human healthy donors, here referred to as DW, BB and DK [23][24][25] . Sequencing data included naïve and memory B-cells of 14   Samples from the validation screen were considered confirmed if -log(EC50) for PrP ≥ 2 (distinct reactivity against PrP) and -log(EC50) for other targets < 2 (no distinct reactivity against any other control target). All data, including the patient-associated demographic and medical data, was stored in a MS-SQL database. Python and R software as well as GraphPad Prism were used for data visualization and statistical testing. Categorical data was tested with chi-square statistics and Bonferroni correction for multiple comparisons was applied. P-values lower than 0.01 were considered significant.

Kappa and Lambda light chain ELISA for total IgG and anti-PrP autoantibodies from plasma
To test total IgG from plasma, 384 well HB plates (Perkin Elmer) were coated with AffiniPure    Same as (f) for NBH and prion to confirm the specific immunoprecipitation and detection of PrP Sc with the peptide P17 from brains of prion-infected mice. +PK: digested with PK; -PK: non-digested with PK.   and DAPI to stain the cell nuclei (in blue). The Fabs detected wtPrP C in the cerebellar granule cell layer (CGL) and molecular layer (ML) of wt and tga20 mice. As expected, the Fabs did not detect wtPrP C in Purkinje cell layer (PCL) of tga20 mice. The higher staining intensity of Fab71 might be caused by its ability to recognize the repetitive epitopes within the OR multiple times.