A transcriptomic signature predicting septic outcome in patients undergoing autologous stem cell transplantation

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.


Highlights :
1) A transcriptomic signature predicts infection in neutropenic patients 2) Expression of eleven genes can be used routinely for sepsis prediction 3) Early anti-infectious treatment may improve sepsis outcome 4) Identification of a predictive transcriptomic signature is ongoing in leukemia *Highlights (for review) 1 Autologous hematopoietic stem cell transplantation (autoHSCT) is a standard treatment in multiple 2 myeloma and relapsed or refractory lymphomas. After autoHSCT, hematologic reconstitution and 3 infectious complications are the main two critical issues. Though many patients develop infectious 4 complications after therapeutic intensification, it remains impossible to predict infection for each 5 individual. The goal of this work was to determine and identify a predictive transcriptomic signature 6 of systemic inflammatory response syndrome (SIRS) and/or sepsis in patients receiving autoHSCT. 7 High throughput transcriptomic and bioinformatics analysis were performed to analyze gene 8 expression modulation in peripheral blood mononuclear cells (PBMCs) in 21 patients undergoing 9 autoHSCT for hematological malignancies (lymphoma or multiple myeloma [MM]). 10 Transcriptomic analysis of PBMCs samples collected just after conditioning regimen identified an 11 eleven genes signature (CHAT, CNN3, ANKRD42, LOC100505725, EDAR, GPAT2, ENST00000390425, 12 MTRM8, C6orf192 and LOC10289230 and XLOC005643) that was able to early predict (at least 2 to 7 13 days before its occurrence) the development of SIRS or sepsis. 14 The possibility of SIRS or sepsis occurrence early prediction (2-7 days before occurrence) opens up to 15 new therapeutic strategies based on preemptive antibiotic and/or antifungal prophylaxis adapted to 16 the specific risk profile of each patient. 17

Introduction
AutoHSCT is based on the administration of myelosuppressive highdose chemotherapy, followed by 23 infusion of autologous hematopoietic stem cells to obtain hematologic reconstitution. Hematopoietic 24 stem cells (HSCs) infusion reduces chemotherapyinduced myelosuppression period and procedure 25 related mortality rate below 3% [1][2][3]. With few exceptions (solid tumors, autoimmune diseases), 26 autoHSCT is essentially indicated for selected hematological malignancies and considered as a 27 standard treatment in young patients with MM and for relapsed or refractory lymphoma. 28 Besides direct toxicity of conditioning regimens, deep (<0.5 G/L neutrophils) and prolonged (usually 29 7-12 days) neutropenia exposes patients to significant risks of infection. The saprophytic gram 30 negative bacilli (such as Escherichia coli) are the most common cause of septic shock [4] and chronic 31 immunosuppression exposes to the risk of fungal infection. An antifungal prophylaxis is usually 32 administered [5] but antimicrobial prophylaxis is less often given because its effectiveness is not 33 clearly established and increases the risk of Clostridium difficile diarrhea [6]. Unfortunately, it is not 34 possible to foresee which patients will develop a SIRS and/or sepsis. Therefore, it remains impossible 35 to adjust the antibiotic or antifungal prophylaxis to the specific risk profile of each patient. 36 The main objective of this work was to determine and identify a predictive transcriptomic signature 37 of the SIRS and/or sepsis in patients receiving autoHSCT, leading to the possibility of a preemptive 38 antiinfectious treatment. 39

41
Patients 42 The prospective study was approved by the institutional review board of the Assistance Publique des 43 Hôpitaux de Marseille (APHM - AORC2012 201208). Written informed consent was obtained from 44 each patient. Patients were admitted in the hematology department of the Conception university 45 hospital for undergoing autoHSCT. All patients were under 65 years and were already followed in 46 the hematology department for MM or highgrade lymphoma. Before autoHSCT, patients were in 47 complete remission (CR) or in partial remission (PR) after conventional chemotherapy. Inclusion 48 criteria were the same as required for being eligible to autoHSCT. Twenty fours patients were 49 included in this study protocol. Twenty one were analyzed. Blood samples were collected for each 50 patient at three moments: before the conditioning regimen (T1), after the conditioning regimen and 51 before the graft infusion (T2) and at the end of the neutropenic phase (T3). 52 All patients (after written informed consent) hospitalized in the Hematology and Cellular Therapy 53 severe clinical symptoms with at least two of the following criteria: a) Temperature higher than 38 °C 61 or lower than 36 °C b) Heart rate higher than 90 beats/min c) Respiratory rate higher than 20 62 breaths/min or PaCO2 lower than 32 mmHg d) White blood cell counts higher than 12,000 cells/mm 3 63 or lower than 4,000 cells/mm 3 , or the presence of more than 10% immature neutrophils. The last 64 criterion cannot be considered in autoHSCT context because of the aplasia phase following HSC 65 with severe sepsis are patients with sepsis and at least one organ dysfunction. Septic choc is defined 67 by severe sepsis associated with refractory hypotension [8]. 68

Conditioning regimens 70
Conditioning regimen for therapeutic intensification was high dose melphalan (200 mg/m 2 ) for 71 patients with MM and BEAM for patients with lymphoma (carmustin 300 mg/m2 at day 6, 72 etoposide 150 mg/m2 from day 5 to 2 twice daily, cytarabin 200mg/m2 from day 5 to 2 twice daily 73 and melphalan 140mg/m2 day 1, with autoHSCT on day 0). The library AgiND is implemented in R software in order to analyze and visualize data. AgiND was 107 developed on Bioconductor library model (tagc.univmrs.fr/ComputationalBiology/AgiND/) and is 108 used to diagnose data quality and datamicroarray normalization. Quantile method was used to 109 normalize data; the objective was to homogenize distribution of microarray intensity [10]. A filter 110 was applied on row data to delete controls, then a second filter was applied to delete genes which 111 were expressed under the background in at least 80% of samples in each group (SIRS - , SIRS+, 112 To test cofactors effects (gender, treatment, infection) on gene expression, GeneANOVA software 114 was used to perform ANalysis Of Variance (ANOVA) on normalized data to determine an estimation 115 Global ANOVA model is given in the following formula: Y = µ + β 1 G + β 2 T + β 3 I + ε, where Y is 117 explained variable, µ is global mean, β 1 , β 2 , β 3 , are model coefficients, and G, T, I are the quantitative 118 variables; β 1 G, is the gene effects, β 2 T, is the treatment effects, β 3 I, the infection effects, and ε is 119 the error term [11]. Differential gene expression analysis was performed using linear Models for 120 Microarray Data (Limma). Limma is a multivariate analysis and takes into account cofactorseffect 121 tested by ANOVA analysis (treatment and gender) ( Table 2 Fluidigm. The preamplification step was followed by an Exonuclease I treatment (BioLegend) to 150 remove unincorporated primers. The final product was diluted 5fold using 18µl of TE Buffer (10 mM 151 TrisHCl, 1mM EDTA). In a second part, 12 x 12 (samples x primers) qRTPCR reactions are performed 152 for each primer pair on each sample on the 12 x 12 array (FlexSix). We used the EvaGreen detection 153 assay for following standard Fluidigm protocols. Ct values were calculated from the system software 154 Biomark RealTime PCR Analysis (Fluidigm). The primers used were: 155

Patient's analysis 159
Twentyfour patients were included in the study, among these patients; 21 validated the molecular 160 criteria for transcriptomic analysis. Among these 21 patients, 6 patients did not develop a fever or 161 SIRS (28.6%), nine patients developed SIRS (42.6%), 5 a sepsis (23.8%) and 1 a severe sepsis (5%). 162 Patients' clinical data are summarized in the Table 1. and no patient had septic shock, thus impeding the possibility to identify a specific transcriptomic 222 signature predictive for these lifethreatening conditions. In addition, since a SIRS/sepsis predictive 223 signature before the conditioning regimen was not identified, the transcriptomic signature was not 224 linked to the patient preauto HSCT status but depended on the conditioning regimen patient's 225

response. 226
We wondered about the definition of sepsis and SIRS and the clustering of sirs and sepsis; 227 first, white blood cells count criteria is not relevant in patients in deep aplasia. Furthermore, in the 228 same way, patients in deep aplasia present grade IIIII anemia especially after BEAM conditioning 229 regimen. In patients with no cardiovascular and/or pulmonary comorbidity, hemoglobin until 8g/dl is 230 tolerated. However, that often results in an increase of the heart rate higher than 90 beats/min 231 and/or respiratory rate higher than 20 breaths/min, especially in patients with fever. 232 Fever is neither sensitive or specific in conventional patients. But, in our severely 233 immunocompromised patients fever is more sensitive and specific than in immunocompetent 234 patients. Nevertheless, it is not unusual to start an antibiotic therapy in patients with no fever but 235 with a microbiological documentation and/or a major increase of the C reactive protein only. 236 At last, we have been very drastic on the definition of our sepsis patients group. Only the patients 237 with fever and a microbiological documentation have been considered in sepsis. Anyway, our 238 immunocompromised patients are paucisymtpomatic and very few infections were clinically 239 probable. 240 Our transcriptomic signature predicts SIRS/sepsis profiles and is more robust than the main 241 confounding factors, such as conditioning regimen, type or gender. BEAM conditioning regimen was 242 more myelosuppressive than melphalan alone and men have had a more significant risk to develop 243 infectious complications than women [13] We also thank Geneviève Victorero and Noushine MossadeghKeller for technical supports. 295    Pangenomic array for the transcriptomic analysis have been used, data were filtered and the controls 360 were suppressed, cofactors effects were tested based on their implication on variation of gene 361 expression. A multivariate analysis - LIMMA - using the cofactors already tested was performed to 362 define the gene differentially expressed using Bioconductor library. The LIMMA model is given in the 363 following formula: Y = α + β 1 . T + β 2 . G + β 3 . I + ε, where Y is the explained variable, α is the global 364 mean, β 1 . T, β 2 . G, β 3 . I, are the model coefficients, T, G, I, are the quantitative variables, β 1 T, is the 365 treatment effect, β 2 . G, is the gene effect, β 3 . I is the infection effect and ε is the error term. Then, we 366 performed multitesting correction by fixing the threshold to 5%. Finally, we adjusted the expression 367 data based on the cofactors before performing the hierarchical clustering *: Linear Models for 368 Microarray data. 369    Ch : Chromosome, +1 : Forward strand, from 3' to 5', -1 : Reverse strand, from 5' to 3'