Portal de Periódicos da CAPES

Carta Acesso aberto Produção Nacional Revisado por pares

BIBTEX RIS

Test trial of spike‐in immunoglobulin heavy‐chain ( IGH ) controls for next generation sequencing quantification of minimal residual disease in acute lymphoblastic leukaemia

2020; Wiley; Volume: 189; Issue: 4 Linguagem: Inglês

10.1111/bjh.16571

1365-2141

Guilherme Navarro Nilo Giusti, Patrícia Yoshioka Jotta, Caroline de Oliveira Lopes, Mônica Aparecida Ganazza, Amilcar Cardoso de Azevedo, Sílvia Regina Brandalise, João Meidânis, José Andres Yunes,

Cancer Genomics and Diagnostics

Minimal residual disease (MRD) is the strongest prognostic factor for acute lymphoblastic leukaemia (ALL). Next generation sequencing (NGS) of immunoglobulin/T-cell receptor (Ig/TCR) repertoires is a promising technology for MRD assessment, overcoming many limitations of current quantitative polymerase chain reaction (qPCR) methods. One challenge for using NGS for MRD consists of normalising the number of reads related to the leukaemia clonotype into its molecule copy number (Knecht et al., 2019). In the present study, we validated in a clinical setting the use of immunoglobulin heavy-chain (IGH) spike-in control plasmids (n = 19) for assessing MRD by NGS (Fig 1A). Repertoire analysis was performed by Vidjil-algo (Duez et al, 2016), and MRD calculations were done using an in-house Python script. We achieved results that were better or comparable to those demonstrated by other published methods, which either did not use spike-in controls (Kotrova et al, 2015; Shin et al, 2017), relied on trade secret methodologies (Faham et al, 2012; Cheng et al, 2018), or did not explore how their systems work (Ladetto et al, 2014; Sekiya et al, 2017; Theunissen et al, 2019). In pilot tests, we observed the need of plasmid linearization before use (Figure S1). Additionally, use of spike-ins at high copy numbers (360–1080) consumed a large portion (55·2%) of the sample's reads (Figure S2). Therefore, subsequent experiments were performed using 19 IGH V(D)J sequences, comprehending 2–3 different clones per VH segment (VH1–7), at final amounts of 10, 40 and 160 copies per reaction. Detailed methods are available as Supplemental Data. This spike-in system was validated in retrospective samples from 110 consecutive patients with B-cell precursor (BCP)-ALL (ethical approval Certificado de Apresentação para Apreciação Ética [CAAE]: 57280616.4.0000.5376) that presented IGH complete rearrangements, as determined in prospective Ig/TCR screening for MRD analysis by qPCR. The numbers of VH-(DH)-JH clonotypes identified at diagnosis (day 0, D0) by conventional MRD screening and NGS are shown in Table I. NGS identified two or more leukaemia clonotypes in more patients (60%) than homo/heteroduplex screening used in the qPCR method (36·4%). Of note, 20 patients presented with two or more leukaemia clonotypes with identical (IgHD)-N-IgHJ regions, indicating clonal evolution. A total of 129 IGH markers had MRD calculated by both qPCR and NGS (Table SIII). Read frequency normalisation into NGS MRD was performed by Family fitting in most cases (86·8%). Read frequencies of spike-in molecules differed across different VH families, likely reflecting variable primer efficiencies (Figure S4). However, MRD by Universal and Family fitting were highly correlated (Figure S5) and equally accurate in comparison to qPCR (Table SIV). Notably, NGS MRD accuracy did not correlate with the number of spike-in reads (Figure S6). Out of the 129 paired qPCR/NGS MRD values, 103 (79·8%) were concordant (Fig 1C): 64 were negative and 39 were positive, strongly correlated (r = 0·82) and accurate [mean absolute error (MAE) in log10 scale was 0·51]. Discordant results (20·2%) occurred mostly in low MRD (10−5) cases (16 qPCR+/NGS− and 10 qPCR−/NGS+; Table 1). For the 15 patients with at least one qPCR+/NGS– clonotype, eight had another IGH marker for analysis: in seven it confirmed MRD negativity (Table SV). Therefore, we believe most of these were truly MRD negative and thus qPCR false-positives (Kotrova et al, 2017). As for the nine patients with qPCR–/NGS+ clonotypes, five had another marker for analysis: three were positive MRD (Table SVI). We believe these results are qPCR false-negatives, as D35 NGS was performed separately from D0, excluding possible cross-contamination. Discordant results were manually curated. One of the 16 qPCR+/NGS− results represented a case of clonal evolution. Manual inspection of D0 NGS data showed three (IgHD)-N-IgHJ-related clonotypes, at 88·8%, 1·2% and 0·7% (Figure S7A). It happened that the more abundant clonotype did not persist at D35, while the other two were not followed, as they were below the identification threshold (5%) at D0. In qPCR, the patient-specific primers fortuitously annealed to all three clonotypes. Therefore, a lower read frequency threshold for leukaemia identification should be considered. Other notable cases are presented in Figure S7B,C. NGS MRD was also calculated without spike-in normalisation and compared to qPCR (Fig 1D). This method resulted in a 78·54 NGS/qPCR ratio, indicating significant scaling errors (spike-in NGS/qPCR = 1·55). As a consequence, use of non-normalised NGS MRD could erroneously attribute high-risk status to many patients. Overestimation of MRD values by non-normalised NGS would probably be more elevated the higher the proportion of non-lymphoid cells in the bone marrow sample. Surprisingly, non-normalised NGS MRD correlated equally well with qPCR (r = 0·81). We concluded that spike-in normalisation does not correct for linearity, but is paramount for scaling adjustments. This spike-in system proved to be reproducible, even with major differences in the spike-in sets used, as patient samples analysed with the pilot and the 19 spike-in set (n = 10) produced similar MRD results (Figure S8). Moreover, NGS MRD presented little variation when calculated using reduced numbers of the spike-in plasmids. Use of only the 160-copies set works well enough (Figure S9), but addition of the 10 copies set is advisable for primer sensitivity assessment. In conclusion, we validated a spike-in system that provides positive controls for every primer used in the assay, allows testing of its sensitivity down to 10−4, and is redundant, as every VH family has representation in more than one spike-in control. Failure of the Family-specific set is safely substituted for MRD calculation by the average of all other spike-in values. We are very thankful to Marc Duez, Mathieu Giraud, Ryan Herbert, Tatiana Rocher, Mikaël Salson, Florian Thonier from the Vidjil Team for help in using and adapting Vidjil for our MRD analysis pipeline. We thank Priscila Pini Zenatti, a researcher at Centro Infantil Boldrini, for technical help with MRD analysis, and Elizabete Delbuono, from Grupo de Apoio ao Adolescente e à Criança com Câncer (GRAACC), for information on the percentage of lymphocyte in D35 bone marrow samples. This work was supported by research funding to José Andrés Yunes from PRONON (Programa Nacional de Apoio à Atenção Oncológica, SIPAR 25000.057709/2015) together with Tetra Park Ltda, Raízen Combustiveis S/A, Globo S.A, Flextronics International Tecnologia Ltda, Águas Guariroba, Antibioticos do Brasil, Renovias Concessionária, Rigesa de Celulose, Empresa Catarinense de Transmissores de Energia, 3M Manaus Ind. Prod. Químicos, STVD Holdings S.A, Bradesco Vida e Previdência, Prolagos S.A, Banco Haitongi, Astra S.A. Indústria e Comércio, Buckman Laboratórios, AEGEA Saneamento e Participações S.A, Rigesa do Nordeste, Rud Correntes, Rico Corretora, Finamax S/A Credito Financ. e Investimento, Japi Indústria e Comércio S.A, Manoel Fernandes Flores, LVE - Locadora de Veículos e Equipamentos Ltda and Danilo Rabetti. GNNG received scholarships from Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq 132978/2017-2) and from Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP 2017/03942-8). José Andrés Yunes received a productivity fellowship from the National Counsel of Technological and Scientific Development (CNPq). João Meidanis acknowledges a grant from FAPESP (2018/00031-7). Guilherme Navarro Nilo Giusti, Patrícia Yoshioka Jotta and José Andrés Yunes conceived and designed the study; Guilherme Navarro Nilo Giusti, Patrícia Yoshioka Jotta and Caroline de Oliveira Lopes performed all sequencing experiments; Sílvia Regina Brandalise and Amilcar Cardoso de Azevedo were responsible for the diagnosis and treatment of patients; Patrícia Yoshioka Jotta, Caroline de Oliveira Lopes and Monica Aparecida Ganazza performed all qPCR experiments; Guilherme Navarro Nilo Giusti, João Meidanis and José Andrés Yunes analysed results; Guilherme Navarro Nilo Giusti and José Andrés Yunes wrote the manuscript. The authors have no conflicting financial interests. Data S1. Supplemental data. Fig S1. Supercoiled and linear spike-in control: MRD estimation comparison. Fig S2. Read fraction and qPCR versus NGS mean absolute error per spike-in subset. Fig S3. Frequency of spike-in reads by total VDJ reads. Fig S4. Spike-in controls amplification per VH family. Fig S5. Universal versus Family/Universal fitting comparison. Fig S6. Accuracy of NGS MRD is not influenced by spike-in read fraction. Fig S7. D0 and D35 leukaemia clones for notable cases. Fig S8. Comparison of MRD values for samples analysed both in the pilot and main studies. Fig S9. Copy number-based subsets of spike-in control produce similar NGS MRD values. Table SI. Sequences of the complete IGH rearrangements used as spike-in control. Table SII. BIOMED-2-based FR2 primers with non-complementary 5ʹ overhangs for complete IGH rearrangements amplification and NGS library construction. Table SIII. Paired qPCR and NGS MRD values for 129 leukaemia markers. Table SIV. Mean Absolute Errors of NGS MRD calculated by Universal or Family/Universal fitting against qPCR MRD. Table SV. qPCR+/NGS− MRD patients results. Table SVI. qPCR−/NGS+ MRD patients results. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.