Characterization of prefusion-F-specific antibodies elicited by natural infection with human metapneumovirus

Human metapneumovirus (hMPV) is a major cause of acute respiratory tract infections in infants and the elderly for which there are no approved vaccines or antibody therapies. The viral fusion (F) glycoprotein is required for entry and is the primary target of neutralizing antibodies, however, little is known about the humoral immune response generated by humans as a result of natural infection. Here, we use stabilized hMPV F proteins to interrogate memory B cells from two elderly donors. We obtained over 700 paired non-IgM antibody sequences representing 563 clonotypes, indicative of a highly polyclonal antibody response to hMPV F in these individuals. Characterization of 136 of these monoclonal antibodies revealed broad recognition of the hMPV F surface, with potent neutralizing antibodies targeting each antigenic site. Cryo-EM structures of two neutralizing antibodies reveal the molecular basis for recognition of two prefusion-specific epitopes at the membrane-distal apex of hMPV F. Collectively these results provide new insights into the humoral response to hMPV infection in the elderly and will guide development of novel vaccine antigens.


INTRODUCTION
Human metapneumovirus (hMPV) is an enveloped negative-sense RNA virus of the 2 Pneumoviridae family that has been circulating for at least 70 years but was only identified in 3 2001 (van den Hoogen et al., 2001). Analysis of hMPV genetic lineages identifies two genotypes, 4 A and B van den Hoogen et al., 2004). The genotypes are further divided into 5 5 subgroups (A1, A2a, A2b, B1 and B2) primarily based upon the attachment protein sequence 6 variability (Nao et al., 2020;van den Hoogen et al., 2004;Yang et al., 2013). hMPV has a 7 seasonality of winter and early spring with co-circulation of the two genotypes frequently 8 observed (Kahn, 2006). It is now recognized as a leading cause of respiratory tract infections in 9 infants with near universal pathogen exposure by the age of five (van den Hoogen et al., 2001;10 Wang et al., 2021). In 2018, it was estimated that the global burden of hMPV-associated acute 11 lower respiratory tract infections in children under five was 14 million cases with approximately 12 643,000 associated hospitalizations (Wang et al., 2021). Despite early exposure, reinfection 13 continues throughout life with symptoms that are generally mild. However, reinfections among 14 the immunocompromised and the elderly can lead to more severe disease such as bronchiolitis 15 and pneumonia Boivin et al., 2007;Walsh et al., 2008). Although vaccines and 16 monoclonal antibodies are in various stages of early clinical development (Biacchesi et al., 2005;17 Chupin et al., 2021;Cox et al., 2014;Cseke et al., 2007;Hamelin et al., 2007;Herfst et al., 2008; 18 Huang et al., 2021;Karron et al., 2018;Levy et al., 2013;Olmedillas et al., 2018;Schuster et al., 19 2015;Stepanova et al., 2020;Stewart-Jones et al., 2021), there are currently no FDA-approved 20 interventions available. To develop an effective hMPV countermeasure, it will be important to 21 4 have a thorough understanding of the humoral immune response that is elicited by natural 22 infection. 23 Like other pneumoviruses, such as respiratory syncytial virus (RSV), the envelope of the 24 virion is decorated with two glycoproteins: the fusion (F) glycoprotein and the attachment (G) 25 glycoprotein. hMPV F and G facilitate viral attachment through interactions with cellular 26 glycosaminoglycans (Chang et al., 2012;Huang et al., 2021;Klimyte et al., 2016;Thammawat et 27 al., 2008). Additionally, F may interact with RGD-binding integrins through a highly conserved 28 RGD motif Cseke et al., 2009;Wei et al., 2014). However, whereas G is 29 dispensable for propagation in cell culture (Biacchesi et al., 2004;Dubois et al., 2019), F is 30 required for viral entry because it facilitates fusion of the viral and host-cell membranes (Mas 31 and Melero, 2013;Melero and Mas, 2015). Although the immune response elicited by hMPV G 32 has been found to be nonprotective, hMPV F elicits a potent neutralizing antibody response 33 (Biacchesi et al., 2004) and is therefore a major focus of vaccine development efforts. 34 Like all class I viral fusion glycoproteins, hMPV F is initially translated as an inactive 35 precursor (F0). Activation requires proteolytic cleavage by a host-cell protease at a conserved 36 RQSR sequence located between the N-terminal F2 subunit and the C-terminal F1 subunit. This 37 cleavage is likely performed by serine proteases such as TMPRSS2 (Shirogane et al., 2008) on the 38 target cell, although some F protein may be cleaved during viral egress. Following proteolysis, 39 the disulfide-linked F2/F1 heterodimers trimerize to adopt the metastable prefusion (preF) 40 conformation (Battles et al., 2017). In response to an unknown stimulus or thermodynamic 41 instability, the preF protein undergoes a dramatic rearrangement of its tertiary structure, 42 extending and inserting the hydrophobic fusion peptide at the N-terminus of the F1 subunit into 43 5 the host-cell membrane. This elongated pre-hairpin intermediate then collapses back upon itself 44 to bring the F1 N-and C-termini, and thus the host-cell and viral membrane, into close proximity, 45 resulting in the fusion of the two membranes and adoption of the highly stable postfusion (postF) 46 conformation . 47 Through large-scale antibody isolation and characterization studies of RSV F, six major 48 antigenic sites (Ø, I, II, III, IV, and V) covering the preF protein surface have been described 49 (Gilman et al., 2016). Antigenic sites Ø and V are preF-specific as they comprise secondary 50 structure elements that undergo significant rearrangement during the preF-to-postF transition 51 (Gilman et al., 2016;McLellan et al., 2013). Antibodies that recognize site III generally have a 52 higher affinity to preF than postF and are thus considered preF-preferring (Goodwin et al., 2018). 53 Sites II and IV are found on both preF and postF (McLellan, 2015;Swanson et al., 2011), whereas 54 site I antibodies have a higher affinity for postF (Gilman et al., 2016). RSV and hMPV F proteins 55 share ~33% sequence identity (van den Hoogen et al., 2002), and structures of the two proteins 56 in their prefusion conformations are similar (Battles et al., 2017;McLellan et al., 2013). Indeed, 57 several previously isolated RSV F-reactive antibodies are cross-reactive with hMPV F, including 58 MPE8, 101F, and M1C7, which recognize sites III, IV, and V, respectively (Corti et al., 2013;Wu et 59 al., 2007;Xiao et al., 2019). Additionally, hMPV F-specific antibodies DS7 and 338 recognize sites 60 I and II (Ulbrandt et al., 2008;Ulbrandt et al., 2006;Wen et al., 2012;Williams et al., 2007), and 61 a human antibody, MPV458, binds hMPV F site Ø within the trimer interface at an epitope that 62 is inaccessible on closed preF trimers (Huang et al., 2020). Although these hMPV F-directed 63 antibodies have helped provide some insight into the antigenicity of hMPV F, there remains a 64 6 need for a comprehensive understanding of the hMPV F-directed antibodies elicited by natural 65 infection. 66 Here, we describe a high-throughput antibody isolation effort using memory B cells from 67 two elderly, naturally hMPV infected donors. From these cells >100 monoclonal antibodies were 68 produced and assessed for reactivity to preF and postF antigens, neutralizing activity, and 69 antigenic site recognition. Furthermore, single-particle cryo-EM structures were obtained for two 70 neutralizing antibodies targeting the preF-specific antigenic sites Ø and V, providing the first 71 structural information on hMPV F antibodies targeting these sites. Collectively, the antibodies 72 isolated and characterized here will help guide development of hMPV F vaccines and 73 immunotherapeutics.

75
Screening and prioritization of donors for antibody discovery 76 Sanofi Pasteur VaxDesign maintains a cohort of PBMC and plasma from human donors for use in 77 pre-clinical research studies. We screened a total of 31 plasma samples from donors between 78 the ages of 59 and 76 under the assumption that these individuals had long histories of repeated 79 exposure to hMPV and were likely to have a robust memory B cell pool for antibody discovery. 80 Donor and donation details can be found in Table S1. Donor plasma were initially tested in an 81 ELISA against hMPV preF A1 strain NL/1/00 (Battles et al., 2017;Stewart-Jones et al., 2021), with 82 titers ranging from 4,000 to 64,000 (Table S1, Figure 1A). We expanded analysis of Donor 2.3 and 83 4.2 to include ELISA titers against hMPV A1 postF (Fig 1B) and determination of virus 84 microneutralization titer against hMPV A2 CAN97-83 (Fig 1C). We also performed flow cytometry 85 analysis on cryopreserved PBMCs from these donors (Fig 1D), through which we observed an 86 7 appreciable frequency of hMPV A1 preF-specific memory B cells. Specifically, we observed that 87 0.50% and 0.42% of class-switched B cells (CD3 -, CD8 -, CD14 -, CD19 + , CD20 + , IgM -) bound the preF 88 probe in Donor 2.3 and Donor 4.2, respectively. Based on these observations, we decided to 89 proceed with single B cell sequencing from these two donors.

90
Identification of hMPV preF-reactive antibodies by memory B cell sequencing and analytics 91 We proceeded with sequencing of memory B cells from Donors 2.3 and 4.2. We sorted a total of 92 8,573 preF-binding B cells from Donor 2.3 and 4,326 pre-F-binding B cells from Donor 4.2 using a 93 fluorescently tagged hMPV preF protein (115-BV-DS) as a probe. The mRNA from individual B 94 cells was bar coded and sequenced, with resulting paired heavy and light chain sequences 95 annotated and analyzed as described in the Methods. We obtained a total of 359 and 381 paired 96 IgG, IgD and IgA sequences from Donors 2.3 and 4.2, respectively (Fig 2A) 100 We next explored IGHV usage in these two donors compared to repertoires of previously 101 published human donors ( Fig. 2B) (Boyd et al., 2010). We found that both donors used IGHV 1-2, 102 3-30, and 5-51 at a higher frequency than the previously published repertoires. Additionally, 103 Donor 2.3 used IGHV 4-30-4 and 4-31 while Donor 4.2 used IGHV 2-5, 3-11, and 3-33 at a higher 104 frequency than previously published donors. Both donors had similar VH nucleotide substitution 105 frequencies with Donor 2.3 having a median of 6.9% and range 0-15.1% and Donor 4.2 having a 106 median of 6.5% and range 0-12.9% (Fig. 2C). Finally, we assessed CDRH3 length in both donors 107 and found that they had a similar distribution, with Donor 2.3 having a majority of 17 to 18 amino-108 8 acid-long CDRH3s and Donor 4.2 having a majority of 16 and 19 amino-acid-long CDRH3s (Fig. 2D Fig. 2A, Table S2). Sequences were selected to obtain diversity in lineage 117 coverage, V gene usage, and CDRH3 length. In addition, some sequences were selected based on 118 CDRH3 features deemed to be of particular interest (e.g., cysteine content). All antibodies were 119 expressed as human IgG1 and characterized by ELISA against preF derived from hMPV subgroups 120 A1(NL/1/00) and B2(TN99-419). In addition, all antibodies were tested for their capacity to 121 neutralize hMPV A2 strain CAN97-83 (Table S3). Of the 136 antibodies expressed, 115 122 demonstrated binding to hMPV F by either ELISA or Octet (84.5% hit rate), and 106 demonstrated 123 binding by ELISA only (titers ≥333 ng/ml were considered non-binding). We sometimes observed 124 disparity between antigen binding by ELISA and Octet, presumably due to differences in antigen 125 presentation on the surfaces. Despite performing the B cell sort with hMPV A1 preF, the vast 126 majority of F-specific antibodies obtained were cross-reactive with A1 and B2 preF (Table S3). 127 Only 1 antibody (SAN27-60) appeared to be A1 specific based on ELISA. Interestingly, 32 of 136 128 antibodies appeared to demonstrate a binding preference for B2 preF based on ELISA binding 129 differential of ≥10-fold. Two of these antibodies, SAN27-45 and SAN27-38, had a differential of 130 9 >100-fold. Of the 115 preF-binding antibodies expressed across both donors, 94 had detectable 131 neutralization (81.7%) at ≤12.5 µg/mL and 14 (12.2%) demonstrated a potent neutralization titer 132 (FRNT50, mAb concentration resulting in 50% reduction in number of foci) of ≤0.1 µg/mL that 133 compared favorably to published antibodies DS7, 338, MPE8, and 101F (Table S3) (Corti et al.,134 2013; Mas et al., 2016;Ulbrandt et al., 2006;Wen et al., 2012). We observed a similar distribution 135 of neutralization potency for mAbs isolated from both donors ( Fig. 3A and 3B To further understand how neutralization potency related to binding specificity, we evaluated 140 mAb neutralization in terms of preF and postF ELISA specificity (Fig. 3C, 3D). For both donors, 141 >75% of hMPV F-binding mAbs were able to bind both pre-and postfusion F to some degree ( Fig.   142 3C). Donor 2.3 had a greater frequency of preF preferring mAbs (i.e., 10x or greater ELISA titer to 143 preF than postF), and all mAbs that were entirely preF specific were also isolated from this Donor. 144 Donor 4.2 had a higher frequency of mAbs with comparable binding to both preF and postF. 145 There was no obvious segregation of neutralization with preF or postF binding preference ( To map the binding sites of the isolated antibodies we conducted a competition assay using 152 biolayer interferometry. Briefly, a prefusion-stabilized hMPV F construct was loaded onto 153 biosensors and saturated with known antibodies. The saturated antigen was then used to test 154 for binding of the isolated hMPV F-specific antibodies. Antibodies DS7 (site I), 338 (site II), MPE8 155 (Site III), and 101F (site IV) were used as the saturating antibodies to map the indicated antigenic 156 sites (Figs. 4, S1). We found that 59 of the 106 F binding antibodies (ELISA) competed with one 157 or more of the four aforementioned antibodies for binding to hMPV preF, whereas 37 of the 106 158 were non-competitive (Figs. S1, S2). These 37 non-competitive antibodies were then competed 159 against each other, resulting in the identification of four additional epitope bins (A, B, C, and D). 160 From the cryo-EM studies described below, Bins B and C were found to be equivalent to the 161 prefusion-specific sites V and Ø, respectively. Bin A was found to be located within the trimer 162 interface, similar to the previously described antibody MPV458, which binds near the apex of the 163 F protein (Fig. S3) (Huang et al., 2020). Bin D antibodies were unable to bind to a preF construct 164 that contained an engineered disulfide bond near the C-terminus (residues 365 and 463) -165 referred to in Methods as DS-CavEs2. Therefore, Bin D was determined to have a membrane-166 proximal epitope near the base of the preF conformation (Figs. 4, S4). 167 The majority of the isolated antibodies could be placed into just 3 of the 13 bins: Sites I/III 168 overlapping, Sites II/III overlapping, and Bin D. Interestingly, we did not observe an obvious 169 segregation of potency to specific antigenic sites. Most bins had at least a few neutralizing 170 antibodies with IC50 <1 µg/mL, with the exception of Bin A and Bin D. This is in contrast to what 171 has been observed for RSV F, where the most potent antibodies primarily target the prefusion-172 specific sites Ø and V, as well as antigenic site IV which is present on both the preF and postF 173 11 conformations (Gilman et al., 2016). For hMPV F, only 1 of the 9 antibodies targeting sites Ø and 174 V had an IC50 < 0.1 µg/mL. The majority (12/14) of the antibodies with IC50 < 0.1 µg/mL competed 175 with antibodies 338, MPE8, and/or 101F, suggesting that the majority of the antibody response 176 to hMPV F as a result of natural infection does not target prefusion-specific epitopes.

177
Cryo-EM studies of two antibodies from unknown competition bins 178 We determined a cryo-EM structure of the SAN27-14 antigen-binding fragment (Fab) from Bin B 179 bound to hMPV F to a global resolution of 3.1 Å. The MPE8 Fab, which has a quaternary epitope, 180 was included in the complex to keep hMPV F in a compact trimer on the grids. Local refinement 181 after subtracting the constant regions of both Fabs led to a better-resolved map of the Fab 182 interfaces for model building (Fig. 5A). The structure revealed that SAN27-14 binds to antigenic 183 site V-in proximity to MPE8 (site III) (Wen et al., 2017)-exclusively via its heavy chain, which 184 buries 726 Å 2 of surface area (Fig. 5B). The 15-amino-acid-long CDRH3 loop, stabilized by a 185 disulfide bond, packs against the hMPV F surface formed by α3 and β3 from the F1 subunit. An 186 N-linked glycan attached to Asn96 of the CDRH3 forms a hydrogen bond with the sidechain of 187 Asn139 on F (Fig. 5C). In addition, the sidechain of CDRH1 Tyr33 forms a hydrogen bond with the 188 sidechain of hMPV F Gln127. hMPV F Leu130, residing in a short loop connecting α2 and α3, is 189 surrounded by Tyr33 and Tyr53 in the CDRH1 and CDRH2 loops, respectively (Fig. 5D). Notably,190 the CDRH2 and subsequent framework region reach toward antigenic site IV on the neighboring 191 hMPV F protomer, establishing a quaternary contact via three salt bridges (Fig. 5E, F). These 192 interactions involve hMPV F residues Lys429 and Glu431 at the hinge region preceding β22, which 193 needs to flip 180° to adopt the postfusion conformation. Collectively, these data demonstrate 194 that SAN27-14 binds to preF at regions in site V and site IV that refold during the pre-to-195 12 postfusion transition. These findings suggest that SAN27-14 neutralizes hMPV by preventing 196 refolding of hMPV F, and thus inhibiting fusion of the viral and host-cell membranes. 197 We also determined a cryo-EM structure of SAN32-2 Fab (Bin C) bound to hMPV F 198 complexed with MPE8 Fab to a global resolution of 3.1 Å (Fig. 6A, Table S5). Local refinement 199 focused on the variable regions of the SAN32-2 Fabs and the membrane-distal half of the F trimer 200 led to a better-resolved map of the Fab interfaces for model building. The structure reveals that 201 SAN32-2 binds to antigenic site Ø, with its heavy and light chains burying 359 Å 2 and 394 Å 2 of 202 surface area, respectively, on hMPV F (Fig. 6B). CDRH2 residues Asp52 and Asp54 form salt 203 bridges with hMPV F Lys179, which resides in the loop between α4 and α5 that rearranges into 204 an extended helix during the preF-to-postF transition (Fig. 6C). CDRH2 residue Arg50 and CDRH3 205 residue Ser98 form hydrogen bonds to Asn180 in this loop as well. The SAN32-2 light chain binds 206 to hMPV F through a network of hydrogen bonds formed between CDRL1 Tyr32 and hMPV F 207 Asp72 as well as between multiple CDRL2 residues and Arg79 of hMPV F (Fig. 6D). Additionally, 208 CDRL3 Asp92 forms a salt bridge with hMPV F Lys68 in the α1 helix.

209
Compared to RSV F, hMPV F has an additional glycosylation site within antigenic site Ø at 210 Asn172 in α4. We had previously suggested that this would create a glycan shield at the apex of 211 hMPV preF and reduce elicitation of site Ø-directed antibodies (Battles et al., 2017). Indeed, only 212 six antibodies were identified as belonging to this competition bin. SAN32-2 is able to evade the 213 glycan shield by binding toward the α1 helix of F2 with a horizontal angle of approach (Fig. 6B). 214 SAN32-2 fits between the Asn57 glycan of the protomer to which it is bound and the Asn172 215 glycan from the neighboring protomer. The SAN32-2 structure thus defines a site Ø epitope on 216 hMPV F and reveals how a neutralizing antibody can evade the glycan shield.

DISCUSSION
Only the fusion protein elicits neutralizing antibodies as a result of hMPV infection (Ryder et al., 219 2010;Skiadopoulos et al., 2006). It is therefore important to understand the antibody response 220 to hMPV F and how its epitopes correlate with neutralization. Although previous studies 221 generated hMPV F-specific antibodies (Bar-Peled et al., 2019;Huang et al., 2020;Mas et al., 2016;222 Ulbrandt et al., 2008;Ulbrandt et al., 2006;Wen et al., 2012;Williams et al., 2007;Wu et al., 223 2007), the mapping of the epitope landscape remained largely incomplete. Initial reports 224 suggested that the majority of the neutralizing response was directed toward epitopes shared on 225 the preF and postF conformations (Battles et al., 2017;Pilaev et al., 2020). In the current study, 226 we successfully utilized a prefusion-stabilized hMPV F probe to isolate and sequence the 227 immunoglobulin heavy and light chain variable regions from circulating B cells of two elderly 228 donors who were naturally infected with hMPV, likely multiple times throughout their lives. We 229 then characterized the preF-binding antibodies using biolayer interferometry and cryo-EM to 230 comprehensively map the antigenic sites of hMPV preF. Our results indicate a broad surface 231 coverage with representative mAbs binding to all six RSV F-analogous antigenic sites (Ø, I, II, III, 232 IV, and V), as well as two novel antigenic sites which have not been previously described (Bin A 233 and Bin D). 234 Unlike what has been observed for the closely related RSV (reviewed in (Ruckwardt et al.,235 2019)), targeting of preF-specific epitopes V and Ø does not appear to be critical to achieve 236 potent neutralization of hMPV. The antigenic sites for the potently neutralizing antibodies 237 described here are diverse and there is no clear segregation to dominant sites. Our panel of 238 antibodies included 14 mAbs with a FRNT50 titer ≤0.1 µg/mL, and these 14 mAbs were found to 239 14 bind to Sites Ø, I, II, III, and IV, as well as hybrid sites II/III and I/III. For RSV, site V has been shown 240 to be the target of very potent neutralizing antibodies (Gilman et al., 2016); however, the three 241 site V antibodies isolated in this study had moderate to low FRNT50 titers between 0.1 and 1 242 µg/mL. 243 Surprisingly, we did not isolate a single antibody that was cross-reactive with RSV F. Prior 244 work has led to the discovery of RSV F/hMPV F cross-reactive mAbs such as MPE8 (Site III, (Corti 245 et al., 2013)), 101F (Site IV, (Wu et al., 2007)), and M1C7 (Site V, (Xiao et al., 2019)). In our work, 246 we isolated five site III-specific mAbs, eight site IV-specific mAbs, and three site V-specific mAbs 247 that were all hMPV specific. All of these had distinct variable gene usage and no obvious 248 sequence convergence with previously published cross-reactive antibodies, which likely explains 249 the absence of RSV cross-reactivity (Xiao et al., 2019). In addition, a number of previously 250 reported cross-reactive antibodies utilized the VH3-21 heavy chain germline and the VL1-40 light 251 chain germline (Corti et al., 2013;Gilman et al., 2016;Xiao et al., 2019). We did observe this 252 pairing in several of the sequences from one subject, but they have not yet been expressed and 253 characterized. 254 We observed diverse heavy chain germline gene usage amongst the preF-sorted B cells, 255 with several interesting deviations from a published, unsorted B cell repertoire (Boyd et al.,256 2010). In particular, we noticed a striking enrichment of IGHV1-2 and IGHV3-30 in the preF-sorted 257 B cells from these donors. The 39 antibodies we expressed to represent these germline variable 258 regions were varied in the antigenic sites recognized as well as neutralization potency, though 259 there was a relatively high frequency of site II/III-specific antibodies from the IGHV3-30 heavy 260 chain germline (7 of 23), and Bin D-specific antibodies from the IGHV1-2 heavy chain germline (7 261 15 of 16). Though these data are only derived from 2 donors, it seems likely that naïve B cells carrying 262 these heavy chain variable regions are disproportionately enriched by hMPV infection. Such 263 information can potentially be leveraged to guide antigen design as well as future antibody 264 discovery campaigns. 265 Notably, we were able to identify two novel hMPV F antigenic sites (Bin A and Bin D). Bin 266 A antibodies did not compete with any of the antibodies used for competition, nor did they 267 compete with our site V and site Ø antibodies. However, a subset within Bin A was shown to 268 partially compete with the previously described trimer-interface-binding antibody MPV458, 269 which binds a linear epitope comprising amino acids 66-87 on the α1 helix ( Fig. S3) (Huang et al.,270 2020). The existence of an antigenic site that resides at least partially within the trimer interface 271 suggests, as previously speculated (Huang et al., 2020), that uncleaved hMPV F on the surface of near the membrane-proximal base of the protein (Fig. S4). Nanobodies recognizing a similar 276 region on RSV F have recently been described (Xun et al., 2021) and are potent neutralizers, 277 whereas Bin D neutralizing antibodies in our study are rare, with only 1 of the 17 having strong 278 neutralization potency (Fig. S1). Further investigation into Bin A and D antibodies may ultimately 279 uncover more potently neutralizing antibodies. 280 The results described here should facilitate development of an effective hMPV vaccine by 281 providing a better understanding of the humoral response to hMPV F-elicited natural infection. 282 Additionally, the antibodies from this study will be useful reagents for future investigation of An hMPV F A1 gene encoding residues 1-490 from strain NL/1/00 with the putative F cleavage 314 site changed to a furin cleavage site (ENPRQSR -> ENPRRRR), as well as substitution of A185P and 315 G294E, was cloned into mammalian expression vector pαH upstream of a C-terminal "GGGS" 316 linker sequence followed by the T4 fibritin trimerization motif "foldon" (Efimov et al., 1994;317 Miroshnikov et al., 1998), an HRV3C protease cleavage site, an 8xHis tag, and a Strep-TagII 318 (Battles et al., 2017). Gibson assembly was used to introduce the prefusion-stabilizing disulfide protein was purified via Strep-Tactin Sepharose resin (IBA) from cell supernatants that had been 326 filtered and buffer-exchanged into PBS by tangential flow filtration. Protein was further purified 327 by size-exclusion chromatography using a Superose 6 Increase 10/300 column (GE Healthcare) in 328 2 mM Tris pH 8.0, 200 mM NaCl, and 0.02% NaN3 running buffer. 329 The prefusion hMPV F construct DS-CavEs2 (hMPV F A1 NL/1/00, residues 1-490), which 330 was used for structural studies, includes the G294E, A185P, and furin cleavage site substitutions 331 as described above as well as disulfides at L110C/N322C, T127C/N153C, A140C/A147C, 332 T365C/V463C and additional substitutions of L219K, V231I, and E453Q (manuscript under 333 review). Expression and purification were as described above. 334 Postfusion hMPV F A1 and B2 constructs were produced as previously described (Mas et 335 al., 2016). Transient transfection of FreeStyle 293F (Thermo Fisher) cells was as described above.  The expressed antibodies, which were selected from the initial Cell Ranger v3 analysis, were 412 reconciled with Cell Ranger 6.0 results using the cell barcode information. As some assembled 413 sequences were considered high-confidence cellular contigs only in Cell Ranger 6.0 (e.g., 414 rearrangements not productive or not cell-associated), some of the antibodies selected for    Tables S4 and S5.

499
Cryo-EM data processing 500 Micrographs were corrected for gain reference and imported into cryoSPARC Live v3.2.0 for initial 501 data processing: motion correction, defocus estimation, micrograph curation, particle picking 502 and extraction, and particle curation through iterative streaming 2D class averaging. 2D averages 503 were used to generate templates and template-based particle picking was carried out. Curated 504 particles were exported to cryoSPARC v3.2 for further processing via rounds of 2D classification, 505 ab initio reconstruction, heterogeneous refinement, and non-uniform homogenous refinement 506 using C3 symmetry. Masking and particle subtraction were used for further non-uniform local 507 refinement. Finally, both global and focused maps were sharpened using DeepEMhancer 508 (Cianfrocco et al.;Sanchez-Garcia et al., 2021) and a combined focused map was generated in 509 PHENIX. EM processing workflows are shown in Figures S5 and S7, and EM validation results are 510 shown in Figures S6 and S8. For model building, an initial hMPV F model was generated from PDB 511 ID: 5WB0 and a Fab model generated by SAbPred server (Dunbar et al., 2016) were used to 512 initially dock into the cryoEM maps using UCSF ChimeraX (Pettersen et al., 2021). Models were 513 built further and iteratively refined using a combination of Coot (Emsley et al., 2010), PHENIX 514 27 (Liebschner et al., 2019), and ISOLDE (Croll, 2018). Model statistics are shown in Tables S4 and   515 S5.    Percentage of antibodies binding to hMPV preF or postF by ELISA. ND indicates antibodies that did not have reportable values for one of the assays required for analysis. The antibodies were defined as "Preferred" if they demonstrated 10-fold or greater ELISA binding titer to one of the protein conformations. The antibodies were defined as "Specific" if the observed binding titer was <333 ng/mL for one format and >333 ng/mL for the other format. D. Neutralization potency of mAbs, with colors reflecting preF/postF specificity.