key: cord-0777357-209cijc7
authors: Galson, Jacob D.; Schaetzle, Sebastian; Bashford-Rogers, Rachael J. M.; Raybould, Matthew I. J.; Kovaltsuk, Aleksandr; Kilpatrick, Gavin J.; Minter, Ralph; Finch, Donna K.; Dias, Jorge; James, Louisa; Thomas, Gavin; Lee, Wing-Yiu Jason; Betley, Jason; Cavlan, Olivia; Leech, Alex; Deane, Charlotte M.; Seoane, Joan; Caldas, Carlos; Pennington, Dan; Pfeffer, Paul; Osbourn, Jane
title: Deep sequencing of B cell receptor repertoires from COVID-19 patients reveals strong convergent immune signatures
date: 2020-05-20
journal: bioRxiv
DOI: 10.1101/2020.05.20.106294
sha: bbc1fb9a060a2438e55531761496cbd2f390ac55
doc_id: 777357
cord_uid: 209cijc7

Deep sequencing of B cell receptor (BCR) heavy chains from a cohort of 19 COVID-19 patients from the UK reveals a stereotypical naive immune response to SARS-CoV-2 which is consistent across patients and may be a positive indicator of disease outcome. Clonal expansion of the B cell memory response is also observed and may be the result of memory bystander effects. There was a strong convergent sequence signature across patients, and we identified 777 clonotypes convergent between at least four of the COVID-19 patients, but not present in healthy controls. A subset of the convergent clonotypes were homologous to known SARS and SARS-CoV-2 spike protein neutralising antibodies. Convergence was also demonstrated across wide geographies by comparison of data sets between patients from UK, USA and China, further validating the disease association and consistency of the stereotypical immune response even at the sequence level. These convergent clonotypes provide a resource to identify potential therapeutic and prophylactic antibodies and demonstrate the potential of BCR profiling as a tool to help understand and predict positive patient responses.

Deep sequencing of B cell receptor (BCR) heavy chains from a cohort of 19 COVID-19 27 patients from the UK reveals a stereotypical naive immune response to SARS-CoV-2 which is 28 consistent across patients and may be a positive indicator of disease outcome. Clonal 29 expansion of the B cell memory response is also observed and may be the result of memory 30 bystander effects. There was a strong convergent sequence signature across patients, and 31

we identified 777 clonotypes convergent between at least four of the COVID-19 patients, 32

but not present in healthy controls. A subset of the convergent clonotypes were 33 homologous to known SARS and SARS-CoV-2 spike protein neutralising antibodies. 34

Convergence was also demonstrated across wide geographies by comparison of data sets 35 between patients from UK, USA and China, further validating the disease association and 36

consistency of the stereotypical immune response even at the sequence level. These 37 convergent clonotypes provide a resource to identify potential therapeutic and prophylactic 38

antibodies and demonstrate the potential of BCR profiling as a tool to help understand and 39

predict positive patient responses. Since the report of the first patients in December 2019 1,2 , the unprecedented global scale of 47 the COVID-19 pandemic has become apparent. The infectious agent, the SARS-CoV-2 48 betacoronavirus 3 , causes mild symptoms in most cases but can cause severe respiratory 49 diseases such as acute respiratory distress syndrome in some individuals. Risk factors for 50

severe disease include age, male gender and underlying co-morbidities 4 . 51

Understanding the immune response to SARS-CoV-2 infection is critical to support 52 the development of therapies. Recombinant monoclonal antibodies derived from analysis of 53 B cell receptor (BCR) repertoires in infected patients or the immunisation of animals have 54 been shown to be effective against several infectious diseases including Ebola virus 5 , rabies 55 6 and respiratory syncytial virus disease 7 . Such therapeutic antibodies have the potential to 56

protect susceptible populations as well as to treat severe established infections. 57

While many vaccine approaches are underway in response to the SARS-CoV-2 58 outbreak, many of these compositions include as immunogens either whole, attenuated 59 virus or whole spike (S) protein -a viral membrane glycoprotein which mediates cell uptake 60 by binding to host angiotensin-converting enzyme 2 (ACE2). The antibody response to such 61 vaccines will be polyclonal in nature and will likely include both neutralising and non-62 neutralising antibodies. It is hoped that the neutralising component will be sufficient to 63

provide long-term SARS-CoV-2 immunity following vaccination, although other potential 64

confounders may exist, such as raising antibodies which mediate antibody-dependent 65 enhancement (ADE) of viral entry 8-10 . While ADE is not proven for SARS-CoV-2, prior studies 66 of SARS-CoV-1 in non-human primates showed that, while some S protein antibodies from 67 human SARS-CoV-1 patients were protective, others enhanced the infection via ADE 11 . An 68 alternative could be to support passive immunity to SARS-CoV-2, by administering one, or a 69 small cocktail of, well-characterised, neutralising antibodies.

Patients recovering from COVID-19 have already been screened to identify 71 neutralising antibodies, following analysis of relatively small numbers (100-500) of antibody 72 sequences 12,13 . A more extensive BCR repertoire analysis was performed on six patients in 73

Stanford, USA with signs and symptoms of COVID-19 who also tested positive for SARS-CoV-74 2 RNA 14 . Although no information was provided on the patient outcomes in that study, the 75 analysis demonstrated preferential expression of a subset of immunoglobulin heavy chain 76

(IGH) V gene segments with relatively little somatic hypermutation and showed evidence of 77 convergent antibodies between patients.

To drive a deeper understanding of the nature of humoral immunity to SARS-CoV-2 79 infection and to identify potential therapeutic antibodies to SARS- increase) in the COVID-19 patients ( Figure 1A ). All of these V gene segments have been 113

previously observed in SARS-CoV-1 or SARS-CoV-2 specific antibodies 16 . IGHV4-34 has been 114

shown to bind both autoantigens 17 and commensal bacteria 18 and has been associated 115

with SLE 19 . Our data extends this, showing that the proportion of sequences containing the 116 autoreactive AVY & NHS sequence motifs within the IGHV region is significantly more 117 frequent in improving COVID-19 patients compared to stable or deteriorating COVID-19 118 patients, specifically in the IGHG1 isotype subclass (p-value = 0.038; Supplementary Figure  119 2). 120

Comparing isotype subclasses showed a significant increase in the relative usage of 121 IGHA1 and IGHG1 in COVID-19 patients ( Figure 1B ) -these are the two first isotype 122

subclasses that are switched to upon activation of IGHM 20 . There was also an increase in 123 the mean CDRH3 length of the BCRs in the COVID-19 patients, that was most pronounced in 124 the IGHA1, IGHA2 and IGHG1 isotype subclasses ( Figure 1C ). 125 126

To further investigate the COVID-19-specific B cell response, we analysed the characteristics 128 of the BCR sequences that are consistent with recent B cell activation -somatic 129 hypermutation, and clonal expansion. In healthy controls, for class-switched sequences, 130

there is a clear unimodal distribution of sequences with different numbers of mutations, 131

and a mean mutation count across IGHA and IGHG isotypes of 17.6 ( Figure 2A ). In the 132 COVID-19 samples, the mean mutation count was 14.4, and there was a bimodal 133 distribution with a separate peak of sequences with no mutations. This bimodal distribution 134 was most pronounced in the IGHG1, IGHG3, and IGHA1 isotype subclasses, corresponding to 135 the increased isotype usages. Such a distribution is consistent with an expansion of recently 136

class-switched B cells that have yet to undergo somatic hypermutation. There was 137 considerable variation between participants in the proportion of unmutated sequences 138

(Supplementary Figure 1) , which had no significant correlation with the number of days 139 since symptom onset (R = 0.09, p = 0.72), but was lower in the deteriorating compared to 140

improving patients ( Figure 2B ) 141

To investigate differential clonal expansion between patients, the Shannon diversity 142

index of each repertoire was calculated (while accounting for differences in read depth 143 through subsampling). A more diverse repertoire is indicative of a greater abundance of 144 different clonal expansions. The BCR repertoires of the COVID-19 patients were significantly 145 more diverse than the BCR repertoires of the healthy controls ( Figure 2C ); this increase in 146 diversity was positively correlated with an increased proportion of unmutated sequences (R 147 = 0.44, p = 0.061; Figure 2D ). Interestingly, when we investigated the largest clonal 148

expansions, despite having a more diverse repertoire, the largest clonal expansions in the 149 COVID-19 samples were larger than in the healthy controls ( Figure 2E ). These large clonal 150 expansions were also highly mutated and had similar levels of mutation between the 151 COVID-19 samples and the healthy controls ( Figure 2F ).

Sequence convergence can be used to identify putative SARS-CoV-2 specific 154 antibodies 155

Given the skewing of the B cell response in the COVID-19 patients to specific IGHV genes, 156

we next investigated whether the same similarity was also seen on the BCR sequence level 157 between different participants. Such convergent BCR signatures have been observed in 158 response to other infectious diseases 21 , and may be used to identify disease-specific 159 antibody sequences.

Of the 435,420 total clonotypes across all the COVID-19 patients, 9,646 (2.2%) were 161

shared between at least two of the participants ( Figure 3A ). As convergence could occur by 162 chance or be due to an unrelated memory response from commonly encountered 163 pathogens, the healthy control dataset was used to subtract irrelevant BCR sequences. Of 164 the 9,646 convergent clonotypes, 1,442 (14.9%) were also present in at least one of the 40 165 healthy control samples. As expected, of the convergent clonotypes that were also present 166

in the healthy control samples, the mean mutation count was significantly greater ( Figure  167 3B), and the mean CDRH3 length significantly shorter ( Figure 3C ) than the convergent 168 clonotypes that were unique to the COVID-19 patients. 169

To identify a set of SARS-CoV-2-specific antibody sequences with high confidence, 170

we identified 777 convergent clonotypes that were shared between at least four of the 171 COVID-19 patients, but not seen in the healthy controls. In parallel, for a comparison of 172 convergent signatures, we performed the same analysis on a cohort of seven metastatic 173 breast cancer patient biopsy samples 22 , which identified 469 convergent clonotypes. These 174 convergent clonotypes were highly specific to each disease cohort ( Figure 3D ). The 777 175

COVID-19 convergent clonotypes had low mutation levels, with a mean mutation count of 2, 176

and only 51 clonotypes with a mean mutation greater than 5. The sequences within the 177 convergent clonotypes were primarily of the IGHG1 (70%) and IGHA1 (16%) subclasses 178

(Supplementary Figure 3A) . The convergent clonotypes used a diversity of IGHV gene 179 segments, with IGHV3-30, IGHV3-30-3 and IGHV3-33 as the most highly represented 180

(Supplementary Figure 3B ). This IGHV gene usage distribution differs between that of the 181 total repertoire, and IGHV3-30 is also the most highly used IGHV gene in the CoV-AbDab 16 . 182

We next tested whether these convergent clonotypes correlated with disease 183 severity. Indeed, 25 of these convergent clonotypes were found to associate with clinical 184 symptoms after correcting for multiple testing, of which 22 were observed at a significantly 185

higher frequency in improving patients ( Figure 3E and Supplementary Figure 4 ). This is a 186 significantly higher proportion associated with clinical symptoms compared to that expected 187 by chance (p-value = 0.018 by random permutations of labels). Interestingly, some of these 188

clonotypes are common in patients comprising >0.1 % of a patient's repertoire. 189

Furthermore, the convergent clonotypes that are associated with clinical symptoms cluster 190 together ( Figure 3F ) and are found primarily in the IGHA1 and IGHG1 isotypes ( Figure 3G ). To further explore whether the convergent clonotypes observed in our study were indeed 195 disease specific, and to determine whether such convergence was common across studies 196

and geographic regions, we compared these 777 convergent clonotypes to public B cell 197

datasets. 198

First, we compared our data to RNAseq data of bronchoalveolar lavage fluid 199 obtained from five of the first infected patients in Wuhan, China 23 . These samples were 200

obtained for the purpose of metagenomic analyses to identify the aetiological agent of the 201 novel coronavirus but were re-analysed to determine whether we could extract any 202 transcripts from BCRs. From the 10,038,758 total reads, we were able to identify 16 unique 203

CDR3 AA sequences ( 204 Supplementary Table 2) . Of these, one had an exact AA match to a sequence in our 205 data and shared the same V gene segment (IGHV3-15), and J gene segment (IGHJ4) usage 206

( Figure 4A) . The sequence had a CDRH3 AA length of 12, so such a match is unlikely to occur 207 due to chance alone. The clonotype that the sequence belonged to contained 699 total 208 sequences and was convergent between 8 of our 19 COVID-19 patients, but not present in 209 the healthy controls. The clonotype was highly diverse, and the sequences had evidence of 210 low mutation from germline, with a mean mutation count over all sequences of 4.8 211

(Supplementary Figure 5) . 212

Next, we compared our 777 convergent clonotypes to CoV-AbDab -the Coronavirus 213

Antibody Database [accessed 10 th May 2020] 16 . At the time of access, this database 214 contained 80 non-redundant CDRH3 sequences from published and patented antibodies 215 proven to bind SARS-CoV-1 and/or SARS-CoV-2. We found 6 of our clonotypes to have high 216 CDRH3 homology to the antibodies in CoV-AbDab ( Figure 4B and Supplementary Figure 6 ). 217

The most striking similarity was to S304, a previously described SARS-CoV-1 and SARS-CoV-2 218 receptor-binding domain antibody able to contribute to viral neutralisation 24 . One of the 219 777 convergent clonotypes contained sequences with an exact CDRH3 AA sequence match 220 and utilised the same IGHV and IGHJ germline gene segments to S304. This clonotype was 221 convergent across 6 patients and had a mean mutation count of 1. We there was also evidence of a proportion of the response arising from memory recall. In the 254 COVID-19 patients, the largest clonal expansions were highly mutated, equivalent to the 255 level observed in healthy control cohort. Such a secondary response to SARS-CoV-2 has 256 been previously observed 26 , and may be due to recall of B cells activated in response to 257

previously circulating human coronaviruses, as recently highlighted 27,28 . 258

We observed a potential relationship between repertoire characteristics and disease 259 state, with improving patients showing a tendency towards a higher proportion of 260 unmutated sequences. The increased prevalence of autoreactive IGHV4-34 sequences in 261

improving COVID-19 patients compared to stable or deteriorating COVID-19 patients 262

potentially suggests a role for natural or autoreactive antibodies in resolving infection and 263

lower risk of pathology. There is a clear need to expand on these findings by using larger 264 sample cohorts and gathering more clinical data to aid understanding of the differences 265 between individuals that respond with mild versus severe disease and have different 266 recovery patterns. Building upon these observations could help to inform the future 267 development of diagnostic assays to monitor and predict the progression of disease in 268 infected patients.

A large number (777) of highly convergent clonotypes unique to COVID-19 were 270

identified. Our approach of subtracting the convergent clonotypes also observed in healthy 271 controls 15 , allowed us to identify convergence specific to the disease cohort. The unbiased 272 nature of the BCR repertoire analysis approach means that, whilst these convergent 273

clonotypes are likely to include many antibodies to the spike protein and other parts of the 274 virus they may also include other protective antibodies, including those to host proteins.

Characterisation of the heavy chains we have identified, coupled with matched light chains 276 to generate functional antibodies will permit analysis of the binding sites and neutralising 277 potential of these antibodies. The report that plasma derived from recently recovered 278 donors with high neutralising antibody titres can improve the outcome of patients with 279

severe disease 29 , supports the hypotheses that intervention with a therapeutic antibody 280 has the potential to be an effective treatment. A manufactured monoclonal antibody or 281 combination of antibodies would also provide a simpler, scalable and safer approach than 282 plasma therapy. 283

Sequence convergence between our 777 convergent clonotypes with heavy chains 284 from published and patented SARS-CoV-1 and SARS-CoV-2 antibodies 16 supports several 285

observations. Firstly, it demonstrates that our approach of finding a convergent sequence 286 signature is a useful method for enriching disease-specific antibodies, as we find matches to 287

known SARS-CoV spike-binding antibodies. Secondly, it shows that the clonotypes observed 288

in response to SARS-CoV-2 overlap with those to SARS- showing that convergence can be seen both between different locations, and different 302 sample types. We believe that the identification of such high BCR sequence convergence 303 between geographically distinct and independent datasets could be highly significant and 304

validates the disease association of the clonotypes, as well as the overall approach. 305

In summary, our BCR repertoire analysis provides information on the specific nature 306 of the B cell response to SARS-CoV-2 infection. The information generated has the potential 307

to facilitate the treatment of COVID-19 by supporting diagnostic approaches to predict the 308 progression of disease, informing vaccine development and enabling the development of 309 therapeutic antibody treatments and prophylactics. 310

Clinical information gathering were collected by members of the direct care team, including duration of symptoms prior to 318 blood sample collection. Current severity was mapped to the WHO Ordinal Scale of Severity. 319

Whether patients at time of sample collection were clinically Improving, Stable or 320

Deteriorating was subjectively determined by the direct clinical team prior to any sample 321

analysis. This determination was primarily made on the basis of whether requirement for 322 supplemental oxygen was increasing, stable, or decreasing comparing current day to 323 previous three days.

Sample collection and initial processing

Blood samples were centrifuged at 150 xg for 15 minutes at room temperature to separate 327 plasma. The cell pellet was resuspended with phosphate-buffered saline (PBS without 328 calcium and magnesium, Sigma) to 20 ml, layered onto 15 ml Ficoll-Paque Plus (GE 329

Healthcare) and then centrifuged at 400 xg for 30 minutes at room temperature without 330 brake. Mononuclear cells (PBMCs) were extracted from the buffy coat and washed twice 331

with PBS at 300 xg for 8 min. PBMCs were counted with Trypan blue (Sigma) and viability of 332 >96% was observed. 5x10 6 PBMCs were resuspended in RLT (Qiagen) and incubated at room 333 temperature for 10 min prior to storage at -80°C. Consecutive donor samples with sufficient 334

RLT samples progressed to RNA preparation and BCR preparation and are included in this 335 manuscript. 336

Metastatic breast cancer biopsy samples were collected and RNA extracted as part 337

of a previously reported cohort 22 . 338 339

Total RNA from 5x10 6 PBMCs was isolated using RNeasy kits (Qiagen). First-strand cDNA was 341 generated from total RNA using SuperScript RT IV (Invitrogen) and IgA and IgG isotype 342 specific primers 30 including UMIs at 50 o C for 45 min (inactivation at 80 o C for 10 min). 343

The resulting cDNA was used as template for High Fidelity PCR amplification (KAPA, 344

Roche) using a set of 6 FR1-specific forward primers 30 including sample-specific barcode 345 sequences (6bp) and a reverse primer specific to the RT primer (initial denaturation at The Immcantation framework (docker container v3.0.0) was used for sequence processing 357 31,32 . Briefly, paired-end reads were joined based on a minimum overlap of 20 nt, and a max 358 error of 0.2, and reads with a mean phred score below 20 were removed. Primer regions, 359

including UMIs and sample barcodes, were then identified within each read, and trimmed. 360

Together, the sample barcode, UMI, and constant region primer were used to assign 361 molecular groupings for each read. Within each grouping, usearch 33 , was used to subdivide 362 the grouping, with a cutoff of 80% nucleotide identity, to account for randomly overlapping 363

UMIs. Each of the resulting groupings is assumed to represent reads arising from a single 364

RNA. Reads within each grouping were then aligned, and a consensus sequence determined. 365

For each processed sequence, IgBlast 34 was used to determine V, D and J gene 366 segments, and locations of the CDRs and FWRs. Isotype was determined based on 367 comparison to germline constant region sequences. Sequences annotated as unproductive 368

by IgBlast were removed. The number of mutations within each sequence was determined 369

using the shazam R package 32 . 370

Sequences were clustered to identify those arising from clonally related B cells; a 371 process termed clonotyping. Sequences from all samples were clustered together to also 372 identify convergent clusters between samples. Clustering was performed using a previously 373 described algorithm 35 . Clustering required identical V and J gene segment usage, identical 374 CDRH3 length, and allowed 1 AA mismatch for every 10 AAs within the CDRH3. Cluster 375 centers were defined as the most common sequence within the cluster. Lineages were 376 reconstructed from clusters using the alakazam R package 36 . The similarity tree of the 377 convergent clonontype CDR3 sequences was generated through a kmer similarity matrix 378 between sequences in R.

Public healthy control data processing

The healthy control BCR sequence dataset used here has been described previously 15 . Only 382 samples from participants aged 10 years or older, and from peripheral blood were used, 383

resulting in a mean age of 28 (range: 11-51). Furthermore, only class-switched sequences 384

were considered.

Public SARS-CoV-2 bronchoalveolar lavage RNAseq data processing

The bronchoalveolar lavage data comes from a previously published study of SARS-CoV-2 388 infection 23 , with data available under the PRJNA605983 BioProject on NCBI. MIXCR v3.0.3 389 was used, with default settings, to extract reads mapping to antibody genes from the total 390

RNASeq data 37 .

Public CoV-AbDab data processing

All public CDRH3 AA sequences associated with published or patented SARS-CoV-1 or SARS-394

CoV-2 binding antibodies were mined from CoV-AbDab 16 , downloaded on 10 th May 2020. A 395 total of 80 non-redundant CDRH3s were identified (100% identity threshold). These 396 sequences were then clustered alongside the representative CDRH3 sequence from each of 397 our 777 convergent clones using CD-HIT 38 , at an 80% sequence identity threshold (allowing 398 at most a CDRH3 length mismatch of 1 AA). Cluster centres containing at least one CoV-399

AbDab CDRH3 and one convergent clone CDRH3 were further investigated.

Public COVID-19 BCR sequence data processing

The fourteen MiSeq "read 1" FASTQ datasets from the six SARS-CoV-2 patients analysed in 403

Nielsen 

Outbreak of pneumonia of unknown etiology in 427

The mystery and the miracle

Clinical features of patients infected with 2019 novel coronavirus in 430 Wuhan

Receptor Recognition by the Novel 432 Coronavirus from Wuhan: an Analysis Based on Decade-Long Structural Studies of 433 SARS Coronavirus

Epidemiological and clinical characteristics of 99 cases of 2019 novel 435 coronavirus pneumonia in Wuhan, China: a descriptive study

A randomized, controlled trial of Ebola virus disease therapeutics

Comparison of a Novel Human Rabies Monoclonal Antibody to 440 Human Rabies Immunoglobulin for Postexposure Prophylaxis: A Phase 2/3

Development of a Humanized Monoclonal Antibody (MEDI-493) 444 with Potent In Vitro and In Vivo Activity against Respiratory Syncytial Virus

Antibody-dependent SARS coronavirus infection is mediated by 447 antibodies against spike proteins

Is COVID-19 receiving ADE from other coronaviruses? Microbes Infect

Immunodominant SARS coronavirus epitopes in humans elicited both 454 enhancing and neutralizing effects on infection in non-human primates

Identification of neutralizing human monoclonal antibodies from 459 Italian Covid-19 convalescent patients

Maturation of naïve and antigen-experienced B-cell receptor 464 repertoires with age

Nucleotide sequence analysis of the V regions of two IgM cold 468 agglutinins: Evidence that the V(H)4-21 gene segment is responsible for the major 469 cross-reactive idiotype

Self-reactive VH4-34-expressing IgG B cells recognizecommensal 471 bacteria

Diversity, cellular origin and autoreactivity of antibody-secreting 473 cell population expansions in acute systemic lupus erythematosus

Lineage tracing of human B cells reveals the in vivo landscape of 476 human antibody class switching

Convergent Antibody Signatures in Human Dengue

The Genomic and Immune Landscapes of Lethal 480 Metastatic Breast Cancer. Cell Rep

A pneumonia outbreak associated with a new coronavirus of probable 482 bat origin

26. Wec, A. Z. et al. Broad sarbecovirus neutralizing antibodies define a key site of 488 vulnerability on the SARS-CoV-2 spike protein

Targets of T cell responses to SARS-CoV-2 coronavirus in humans with 491 COVID-19 disease and unexposed individuals

Pre-existing and de novo humoral immunity to SARS-CoV-2 in humans

Effectiveness of convalescent plasma therapy in severe COVID-19 496 patients

Design and standardization of PCR primers and protocols 498 for detection of clonal immunoglobulin and T-cell receptor gene recombinations in 499 suspect lymphoproliferations: report of the BIOMED-2 Concerted Action BMH4-CT98-500 3936

pRESTO: a toolkit for processing high-throughput 502 sequencing raw reads of lymphocyte receptor repertoires

Change-O: A toolkit for analyzing large-scale B cell immunoglobulin 505 repertoire sequencing data

Search and clustering orders of magnitude faster than BLAST

Supplementary Figure 1. Distribution of sequences with different numbers of mutations from germline

COVID-19 BCR sequence data will be made available upon publication. IGHV2-26  IGHJ3  CARDSGRHLGPFDIW  IGHV1-2  IGHJ3  CATPYYYDGGLDAFDIW  IGHV3-74  IGHJ5  CARDLSRTNWFDPW  IGHV3-15  IGHJ4  CTTDLHDYGDSDYW  IGHV3-15  IGHJ4  CTTDFGGMITFGGVLRRI  IGHV3-21  IGHJ4  CARAQSRGGYDSFFDFW  IGHV3-21  IGHJ4  CGRGGPGTGIDYW  IGHV4-59  IGHJ5