PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of plosonePLoS OneView this ArticleSubmit to PLoSGet E-mail AlertsContact UsPublic Library of Science (PLoS)
 
PLoS One. 2017; 12(3): e0173532.
Published online 2017 March 9. doi:  10.1371/journal.pone.0173532
PMCID: PMC5344481

A BCR-ABL1 cutoff of 1.5% at 3 months, determined by the GeneXpert system, predicts an optimal response in patients with chronic myeloid leukemia

Matthaios Speletas, Editor

Abstract

In chronic myeloid leukemia (CML) patients, 3-month BCR-ABL1 levels have consistently been correlated with further outcomes. Monitoring molecular responses in CML using the GeneXpert (Cepheid) platform has shown an optimal correlation with standardized RQ-PCR (IS) when measuring BCR-ABL1 levels lower than 10%, as it is not accurate for values over 10%. The aim of the present study was to determine the predictive molecular value at three months on different outcome variables using the Xpert BCR-ABL1 MonitorTM assay (Xpert BCR-ABL1). We monitored 125 newly diagnosed consecutive CML patients in the chronic phase (CML-CP) using an automated method: Xpert BCR-ABL1. Only 5% of patients did not achieve an optimal response at 3 months, and the 10% BCR-ABL1 cutoff defined by RQ-PCR (IS) methods was unable to identify significant differences in the probabilities of achieving a complete cytogenetic response (CCyR) (50% vs. 87%, p = 0.1) or a major molecular response (MMR) (60% vs. 80%, p = 0.29) by 12 months. In contrast, a cutoff of 1.5% more accurately identified differences in the probabilities of achieving CCyR (98% vs. 54%, p<0.001) and MMR (88% vs. 56%, p<0.001) by 12 months, as well as probabilities of treatment changes (p = 0.005). Therefore, when using the Xpert BCR-ABL1 assay, a cutoff of 1.5% at 3 months could with high probability identify patients able to achieve an optimal response at 12 months.

Introduction

Over the last decade, with the incorporation of tyrosine kinase inhibitors (TKIs) for the treatment of chronic myeloid leukemia (CML), the expected overall survival of these patients has reached that of the general population at the same age interval [1]. However, some patients do not respond favorably to TKIs, and the probabilities of progression and death remain high in this patient subset [2]. Thus, the early identification of this group of patients is crucial for determining treatment changes [3]. Real-time quantitative PCR (RQ-PCR) is the molecular procedure recommended by a panel of experts for monitoring CML patients [4]. Nevertheless, conventional RQ-PCR analysis of BCR-ABL1 transcripts is a labor-intensive procedure, incorporating numerous pre- and post-analytical phases, all of which are potential sources of variation and require rigorous technical attention [5]. In recent years, there has been increased effort concerning the standardization of molecular monitoring, and the International Scale (IS) for BCR-ABL1 RQ-PCR quantification has been widely adopted to mitigate the effects of variability between laboratories.

Unfortunately, these initiatives do not address all aspects of the testing process, leading to variable results [6]. Thus, new automated procedures have been incorporated to simplify the process and time spent on molecular monitoring. The Xpert BCR-ABL1 PCR system, performed using a GeneXpert automated analyzer, is a cartridge-based assay with minimal pre-analytical and analytical states enabling faster and easier BCR-ABL1 monitoring than RQ-PCR [7]. Our previous studies, as well as those of other groups, have shown a good correlation between Xpert BCR-ABL1 and conventional RQ-PCR for values under 10% [810]. Similar results were observed by Enjeti et al. [11] and O’Dwyer et al. [12] in prospective studies comparing these two methods.

A cutoff of 10% at 3 months using conventional RQ-PCR (IS) has consistently been correlated with clinically relevant outcomes, including the probabilities of achieving a complete cytogenetic response (CCyR), major molecular response (MMR), progression-free survival (PFS) and OS, in patients treated with frontline imatinib and a second-generation TKI (2GTKI) [1318]. Thus, NCCN and ESMO guidelines have recommended changes for the treatment of patients who do not achieve this milestone at 3 months [3,19]. Conversely, ELN recommendations do not consider this group of patients as failures but as patients that should be closely monitored, reflecting the high risk of subsequently fulfilling failure criteria during follow-up [20].

The aim of the present study was to validate the predictive value on different outcome variables of molecular responses at 3 months using the Xpert BCR-ABL1 assay, including outcomes defined as “optimal” by ELN recommendations such as achieving MMR and CCyR at 12 months [20].

Methods

Patients and methods

We examined 125 new consecutive CML chronic phase (CML-CP) patients diagnosed and followed up in 13 Spanish centers between 2010 and 2013. All patients were Philadelphia chromosome-positive with confirmed BCR-ABL1 rearrangement (e13a2 or e14a2 transcripts) based on qualitative PCR at presentation. The baseline characteristics of the patients are shown in Table 1. The present study was approved by the Hospital Universitario Ramón y Cajal Research ethics committee. All clinic hematologic data from each patient was collected after written inform consent, approved by the ethic committee, was obtained. Signed informed consents were recorded within patient's clinical history.

Table 1
Baseline characteristics.

The monitoring and treatment strategies were selected at the discretion of the hematologist, thereby reflecting the actual treatment of CML patients in clinical practice. The first-line treatments consisted of imatinib (IM), nilotinib (NI), dasatinib (DA) and bosutinib (BO) in 58% (73), 28% (34), 13% (17), and 1% (1) of the patients, respectively.

BCR-ABL1 transcript quantification was performed using the automated method Xpert BCR-ABL1 Monitor™, Cepheid, aligned to the 0.1% BCR-ABL1 ratio according to the standards of the World Health Organization [7]. The samples were analyzed according to the manufacturer’s recommendations.

Evaluation of the responses

Cytogenetic responses were defined according to ELN definitions. Particularly, CCyR was defined as the absence of detectable Philadelphia chromosome in 20 mitotic cells and/or ≤1% BCR-ABL1 cells based on the interphase FISH analysis of 200 nucleated cells.

A molecular response was described as MMR when BCR-ABL1/ABL1 transcript levels were ≤0.1% (IS), whereas MR4.0 was defined as either detectable disease ≤0.01% BCR-ABL1 (IS) or undetectable BCR-ABL1 with a detection limit of 0.01%. MR4.5 was defined as either detectable disease ≤0.0032% BCR-ABL1 (IS) or undetectable BCR-ABL1 with a detection limit of 0.0032%, and MR5.0 was defined as either detectable disease ≤0.001% BCR-ABL1 (IS) or undetectable BCR-ABL1 with a detection limit of 0.001%.

An event was defined as the loss of CCyR or complete hematological response (CHR), progression to accelerated or blastic phase, need for treatment change or death for any reason.

Statistical methods

Data on continuous variables are presented as medians and quartiles. Categorical variables were analyzed based on absolute and relative frequencies.

The proportions of patients who achieved MMR and CCyR after first-line treatment for 1 year and the response at 3 months were compared by applying Pearson’s chi-square test or Fisher’s exact test, depending on the expected frequency counts of the cells. All analyses were performed on an intention-to-treat basis unless otherwise stated. A p-value <0.05 was considered significant, and the contrasts were bilateral.

To determine the best cutoff for the BCR-ABL1/ABL1 level using Xpert BCR-ABL1 at 3 months to predict CCyR and MMR at one year, a receiver-operating characteristic (ROC) curve was generated. The same methodology was followed as previously described but applying the newly obtained threshold.

To identify potential covariates, multivariate analyses were performed using logistic regression to predict CCyR and MMR after 1 year, introducing the classification of the patients according to the new threshold at 3 months together with age, gender and Sokal score.

Results

Outcomes and responses

The median follow-up for the entire cohort was 43 months (range: 14–61). The overall probability of achieving CCyR and MMR at 12 months was 84% (106/125) and 68% (86/125), respectively. Table 2 shows the CCyR and MMR probabilities for each specific TKI used in first-line therapy. Twenty-nine patients (23%) required treatment changes, reflecting resistance (n = 12) or intolerance (n = 17). Among all patients receiving frontline treatment with imatinib, 23 of 73 patients received 2GTKI (dasatinib or nilotinib) as second-line treatment. Six of the 52 patients treated with 2GTKI (nilotinib, dasatinib or bosutinib) frontline therapy required treatment changes: 3 patients were changed to imatinib (2 patients showed intolerance and 1 patients showed primary resistance), 1 patient received ponatinib (reflecting primary resistance), and 1 patient received stem cell transplantation. Mutational analysis was performed in patients with failure criteria by ELN recommendations at any time point, nevertheless it was not performed in neither of the six patients who did not achieve a BCR-ABL1 levels below 10% at 3 months. Twenty-nine percent (37/125) of the patients suffered an event during follow-up, but only 2 patients progressed to blast crisis (1.6%). Eight (6.4%) patients died (6 patients died of causes not related to CML) during the follow-up period. The two patients who progressed to blast crisis died, both patients were treated with imatinib first line and none of them achieved an optimal response by 3 months. All of the six patients who died of causes not related to CML achieved an optimal response by 6 months.

Table 2
Complete cytogenetic responses and major molecular response probabilities for each specific TKI used in first-line therapy.

Three-month response and its correlation with subsequent outcomes

At 3 months, molecular assessments were performed in 112 patients. No patients died or progressed during the first 3 months of follow-up.

We attempted to validate the 10% cutoff using the Xpert BCR-ABL1 method as a surrogate endpoint to predict subsequent outcomes. Only 6/112 (5%) patients had BCR-ABL1 levels >10% at 3 months, a value significantly lower than historical series when using standardized RQ-PCR. All patients treated with 2GTKI achieved BCR-ABL1 levels ≤10% at 3 months. The molecular responses at 3 months after specific TKI therapy are shown in Table 3.

Table 3
Molecular responses for each specific TKI at 3 months.

We observed that when using the Xpert BCR-ABL1 assay, the 10% cutoff at 3 months was not associated with the probability of achieving CCyR (50% vs. 87%, p = 0.1) or MMR (60% vs. 80%, p = 0.29) by 12 months (Table 4). To identify a BCR-ABL1 level correlated with the achievement of MMR and CCyR at 12 months, we used an ROC curve to identify the optimal cutoff that would facilitate the classification of patients as high or low risk with maximal sensitivity and specificity. The area under the curve (AUC) of the ROC curve was 0.88. At 3 months, patients with transcript levels below 1.5% showed significantly better probabilities of CCyR response at 12 months, with 81% and 94% sensitivity and specificity, respectively, for CCyR. With this new cutoff, the probabilities for CCyR and MMR at 12 months were 98% vs. 54% (p<0.001) and 88% vs. 56% (p<0.001), respectively (Table 5). These results were also corroborated in the group of patients treated exclusively with imatinib after excluding those treated with frontline 2GTKI (p = 0.003). The probability of treatment changes was also greater for patients with a molecular response > 1.5% at 3 months (42.9% vs. 16.9%, p = 0.005), and this cutoff also identified the probability of achieving MMR or stronger responses at any time (100% vs. 85%, p = 0.002).

Table 4
Probabilities of complete cytogenetic response and major molecular response at 12 months according to the molecular response (≤10%) at 3 months.
Table 5
Probabilities of complete cytogenetic responses and major molecular responses at 12 months according to the molecular response (≤1.5%) at 3 months.

Using MMR with a 12-month endpoint, we generated a multivariate regression model that included the variables listed in Table 1 together with the molecular response at 3 months (cutoff of >1.5). The effect of the molecular response at 3 months on MMR at 12 months was not affected through other variables (OR = 68.9; 95% confidence interval: 8.4–568.3, p<0.001).

Validation of the 1.5% cutoff in an independent cohort of CML patients treated with frontline 2GTKI

Because most of the patients in the study cohort were treated with frontline imatinib, we validated the observed results in a cohort of CML patients treated with 2GTKI as first-line therapy. A total of 57 consecutive patients from Andalusian CML Group Registry were examined with a median follow-up of 38 months (3–56). All patients were monitored using the Xpert BCR-ABL1 assay. The median age was 48 years (18–74). The ratio of men to women was 59/41, and the risk groups according to Sokal Score were 48%, 30% and 22% for low, intermediate and high risk, respectively. First-line treatment consisted of nilotinib and dasatinib in 58% and 42% of patients, respectively. Overall, the probability of achieving CCyR and MMR at 12 months was 92% (48/52) and 82% (39/47), respectively. Ten patients (17%) required treatment changes as a result of resistance (n = 3), not achieving MMR (n = 3) or intolerance (n = 4). No patients progressed to advanced phases, and only 1 patient died during follow-up (not CML related). The overall median value of BCR-ABL1 at 3 months was 0.16%.

Consistent with the original cohort of patients treated with first-line 2GTKI, all patients achieved a BCR-ABL1 level ≤10% at 3 months; therefore, this threshold did not predict further evolution of the disease. We classified this new cohort based on the new cutoff observed in the primary population (BCR-ABL1 level at 3 months ≤1.5%); 77% of patients achieved BCR-ABL1 levels ≤1.5%, whereas 23% of patients did not. This cutoff also predicted the probability to obtain MMR by 12 months (91% vs. 44% (p = 0.025).

Discussion

The degree of molecular response before reaching CCyR was not relevant until some years ago, when several groups showed that a new threshold of 10% at 3 months identified a group of patients with different probabilities of achieving CCyR, MMR, and MR4.5 as well as PFS and OS [1317]. Therefore, achieving a molecular response ≤10% at 3 months has been included in the guidelines as a new surrogate endpoint that should be reached to maximize favorable outcomes [3,19,20]. Nonetheless, this study is the first to examine the association between molecular responses at 3 months using the Xpert BCR-ABL1 assay.

Automated molecular platforms, such as GeneXpert, are currently used in many centers as the only molecular procedure for monitoring CML patients. This increased use primarily reflects the simplicity and rapidity of the automated procedure compared with standardized RQ-PCR. As previously described, our studies and those of others have reported a good correlation between automated procedures and standardized RQ-PCR for monitoring patients once CCyR has been achieved. Regardless, a good correlation has not been observed in patients with BCR-ABL1 levels >10% [812].

In our study, we showed how a 10% BCR-ABL1 threshold at 3 months, measured using the Xpert BCR-ABL1 method, does not predict CCyR or MMR. We propose that this discrepancy is due to the low number of patients with BCR-ABL1 levels above 10% and likely reflects the overestimation of molecular responses in patients without CCyR using the Xpert assay. We observed that nearly 94% of the patients treated with imatinib obtained an optimal response (≤10%) at 3 months. Moreover, all patients treated with 2GTKI in the original and validation cohorts achieved BCR-ABL1 levels of ≤10% at 3 months.

We determined a new cutoff using an ROC curve; this new cutoff of 1.5% predicted the probabilities of achieving MMR and CCyR at 12 months and also the probabilities for treatment changes and deep molecular responses when BCR-ABL1 levels are measured using the Xpert method. Moreover, this new threshold was observed as the only factor for predicting responses at 12 months compared with classical risk factors, including sex, age and even the Sokal risk index.

The significantly lower rate of MMR at 12 moths for patients with BCR-ABL1 values>1.5% at 3 months is consistent with the reports by Branford et al. [15] and Hughes et al. [18]. In these studies, the predictive threshold at 3 months, by RQ-PCR (IS), was found at 1.45% and 1%, respectively, values that are very similar to those obtained in the present study using the Xpert BCR-ABL1 assay.

There were no differences in progressions or OS, likely reflecting the low number of events. In the cohort examined in the present study, most of the patients (58%) were treated with imatinib first-line therapy, and the results were similar when considering only this subgroup of patients.

Currently, a significant percentage of patients are treated with first-line 2GTKI; therefore, we validated the value of the new cutoff using the Xpert system in an independent cohort of patients treated with frontline 2GTKI therapy.

The main limitation of this study was the lack of simultaneous monitoring with a BCR-ABL1 (IS) EUTOS method, which should be addressed in a future study.

In conclusion, we do not question the 10% cutoff at three months analyzed by the standard RQ-PCR method. However, we did show that when using the current version of Xpert BCR-ABL1, a cutoff of 1.5% at 3 months can better identify patients with different probabilities of achieving an optimal response at 12 months. This threshold was validated in an independent sample, and the finding provides a basis for future, larger studies of the use of this practical and widespread method.

Acknowledgments

This work was partially supported by grants from the Asociación Malagueña para la Investigación en Leucemia (AMPILE).

Contributors

Members of the Spanish Group of Chronic Myeloid Leukemia (GELMC) are: Valentín García-Gutiérrez (Hematology Department, Hospital Universitario Ramón y Cajal, Madrid, Spain). María T. Gómez-Casares (Hematology Department, Hospital Universitario Doctor Negrín, Las Palmas de Gran Canaria, Spain). José M. Puerta (Unidad de Gestión Clínica Hematología y Hemoterapia, Hospital Universitario Virgen de las Nieves, Granada, Spain). Juan M. Alonso-Domínguez (Hospital Universitario Fundación Jiménez Díaz, Madrid, Spain). Santiago Osorio (Hematology Department, Hospital Universitario Gregorio Marañón, Madrid, Spain). Juan C. Hernández-Boluda (Hematology Department, Hospital Clínico, Valencia, Spain). Rosa Collado (Genetic Laboratory, Hematology Department, Consorcio Hospital General Universitario, Valencia, Spain). María J. Ramírez (Hematology Department, Hospital de Jerez de la Frontera, Jerez de la Frontera, Cádiz, Spain). Fátima Ibáñez (Hematology Department, Hospital San Pedro de Alcántara, Cáceres, Spain). Alicia Rodríguez (Hematology Department, Hospital Universitario Virgen Macarena, Sevilla, Spain). Concepción Ruiz-Nuño and Antonio Jiménez-Velasco (Hematology Department, Hospital Regional Universitario de Málaga, IBIMA, Spain). Juan L. Steegmann (Hematology Department & IIS-IP. Hospital Universitario de la Princesa, Madrid, Spain. Lead author of the GELMC. moc.liamg@nnamgeetslj).

Funding Statement

This work was partially supported by grants from the Asociación Malagueña para la Investigación en Leucemia (AMPILE). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability

Data Availability

All relevant data are within the paper.

References

1. Pfirrmann M, Baccarani M, Saussele S, Guilhot J, Cervantes F, Ossenkoppele G, et al. Prognosis of long-term survival considering disease-specific death in patients with chronic myeloid leukemia. Leukemia. 2016;30(1):48–56. doi: 10.1038/leu.2015.261 [PubMed]
2. Pearson E, McGarry L, Gala S, Nieset C, Nanavaty M, Mwamburi M, et al. Disease-related mortality exceeds treatment-related mortality in patients with chronic myeloid leukemia on second-line or later therapy. Leuk Res. 2016;43:1–8. doi: 10.1016/j.leukres.2016.02.001 [PubMed]
3. Baccarani M, Pileri S, Steegmann JL, Muller M, Soverini S, Dreyling M. Chronic myeloid leukemia: ESMO clinical practice guidelines for diagnosis, treatment and follow-up. Ann Oncol. 2012;23(Suppl 7):vii72–7. [PubMed]
4. Cross NCP, White HE, Müller MC, Saglio G, Hochhaus A. Standardized definitions of molecular response in chronic myeloid leukemia. Leukemia. 2012;26(10):2172–5. doi: 10.1038/leu.2012.104 [PubMed]
5. Foroni L, Wilson G, Gerrard G, Mason J, Grimwade D, White HE, et al. Guidelines for the measurement of BCR-ABL1 transcripts in chronic myeloid leukaemia. Br J Haematol. 2011;153(2):179–90. doi: 10.1111/j.1365-2141.2011.08603.x [PubMed]
6. Branford S, Fletcher L, Cross NCP, Müller MC, Hochhaus A, Kim DW, et al. Desirable performance characteristics for BCR-ABL measurement on an international reporting scale to allow consistent interpretation of individual patient response and comparison of response rates between clinical trials. Blood. 2008;112(8):3330–8. doi: 10.1182/blood-2008-04-150680 [PubMed]
7. McNiven M, Talaulikar D. Establishment of a conversion factor for the Cepheid GeneXpert BCR-ABL assay. Pathology. 2012;44(1):55–7. doi: 10.1097/PAT.0b013e32834e4203 [PubMed]
8. Dufresne SD, Belloni DR, Levy NB, Tsongalis GJ. Quantitative assessment of the BCR-ABL transcript using the cepheid Xpert BCR-ABL Monitor assay. Arch Pathol Lab Med. 2007;131(6):947–50. doi: 10.1043/1543-2165(2007)131[947:QAOTBT]2.0.CO;2 [PubMed]
9. Cayuela JM, Macintyre E, Darlington M, Abdelali R Ben, Fund X, Villarese P, et al. Cartridge-based automated BCR-ABl1 mRNA quantification: Solving the issues of standardization, at what cost? Haematologica. 2011;96(5):664–71. doi: 10.3324/haematol.2010.034389 [PubMed]
10. López-Jorge CE, Gómez-Casares MT, Jiménez-Velasco A, García-Bello MA, Barrios M, Lopez J, et al. Comparative study of BCR-ABL1 quantification: Xpert assay, a feasible solution to standardization concerns. Ann Hematol. 2012;91(8):1245–50. doi: 10.1007/s00277-012-1468-4 [PubMed]
11. Enjeti A, Granter N, Ashraf A, Fletcher L, Branford S, Rowlings P, et al. A longitudinal evaluation of performance of automated BCR-ABL1 quantitation using cartridge-based detection system. Pathology. 2015;47(6):570–4. doi: 10.1097/PAT.0000000000000293 [PMC free article] [PubMed]
12. O’Dwyer ME, Swords R, Nagler A, McMullin MF, le Coutre PD, Langabeer SE, et al. Nilotinib 300mg BID as frontline treatment of CML: Prospective analysis of the Xpert BCR-ABL Monitor system and significance of 3-month molecular response. Leuk Res. 2014;38(3):310–5. doi: 10.1016/j.leukres.2013.11.016 [PubMed]
13. Hanfstein B, Muller MC, Hehlmann R, Erben P, Lauseker M, Fabarius A, et al. Early molecular and cytogenetic response is predictive for long-term progression-free and overall survival in chronic myeloid leukemia (CML). Leukemia. 2012;26(9):2096–102. doi: 10.1038/leu.2012.85 [PubMed]
14. Jabbour E, Kantarjian HM, Saglio G, Steegmann JL, Shah NP, Boqué C, et al. Early response with dasatinib or imatinib in chronic myeloid leukemia: 3-year follow-up from a randomized phase 3 trial (DASISION). Blood. 2014;123(4):494–500. doi: 10.1182/blood-2013-06-511592 [PubMed]
15. Branford S, Kim DW, Soverini S et al. Initial molecular response at 3 months may predict both response and event-free survival at 24 months in imatinib-resistant or -intolerant patients with Philadelphia chromosome-positive chronic myeloid leukemia in chronic phase treated with nilotinib. J Clin Oncol. 2012;30(35):4323–4329. doi: 10.1200/JCO.2011.40.5217 [PMC free article] [PubMed]
16. Brümmendorf TH, Cortes JE, de Souza CA, Guilhot F, Duvilli?? L, Pavlov D, et al. Bosutinib versus imatinib in newly diagnosed chronic-phase chronic myeloid leukaemia: Results from the 24-month follow-up of the BELA trial. Br J Haematol. 2015;168(1):69–81. doi: 10.1111/bjh.13108 [PMC free article] [PubMed]
17. Jain P, Kantarjian H, Nazha A, O’Brien S, Jabbour E, Romo CG, et al. Early responses predict better outcomes in patients with newly diagnosed chronic myeloid leukemia: results with four tyrosine kinase inhibitor modalities. Blood. 2013;121(24):4867–74. doi: 10.1182/blood-2013-03-490128 [PubMed]
18. Hughes TP, Saglio G, Kantarjian HM, Guilhot F, Niederwieser D, Rosti G, et al. Early molecular response predicts outcomes in patients with chronic myeloid leukemia in chronic phase treated with frontline nilotinib or imatinib. Blood. 2014;123(9):1353–60. doi: 10.1182/blood-2013-06-510396 [PubMed]
19. O’Brien S, Radich JP, Abboud CN, Akhtari M, Altman JK, Berman E, et al. Chronic Myelogenous Leukemia, Version 1.2014. J Natl Compr Canc Netw [Internet]. 2013;11(11):1327–40. Available from: http://www.ncbi.nlm.nih.gov/pubmed/24225967 [PMC free article] [PubMed]
20. Baccarani M, Deininger MW, Rosti G, Hochhaus A, Soverini S, Apperley JF, et al. European LeukemiaNet recommendations for the management of chronic myeloid leukemia: 2013. Vol. 122, Blood. 2013. p. 872–84. doi: 10.1182/blood-2013-05-501569 [PubMed]

Articles from PLoS ONE are provided here courtesy of Public Library of Science