Research ArticleCancer

High-throughput sequencing of the T cell receptor β gene identifies aggressive early-stage mycosis fungoides

See allHide authors and affiliations

Science Translational Medicine  09 May 2018:
Vol. 10, Issue 440, eaar5894
DOI: 10.1126/scitranslmed.aar5894

Predicting progression in a T cell lymphoma

Mycosis fungoides (MF) is an often indolent cutaneous T cell lymphoma identifiable by the T cell receptor gene TCRB in skin biopsies. de Masson et al. performed high-throughput sequencing of MF skin lesions to determine whether the frequency of the most abundant TCRB sequence could predict the fraction of cases that progress to more aggressive disease. This tumor clone frequency (TCF) outperformed commonly used prognostic indicators of disease progression and overall survival in MF, and was particularly useful in early-stage disease. The TCF may thus help identify which MF patients in the same clinical disease stage are in fact more likely to progress to life-threatening disease.

Abstract

Mycosis fungoides (MF), the most common cutaneous T cell lymphoma (CTCL) is a malignancy of skin-tropic memory T cells. Most MF cases present as early stage (stage I A/B, limited to the skin), and these patients typically have a chronic, indolent clinical course. However, a small subset of early-stage cases develop progressive and fatal disease. Because outcomes can be so different, early identification of this high-risk population is an urgent unmet clinical need. We evaluated the use of next-generation high-throughput DNA sequencing of the T cell receptor β gene (TCRB) in lesional skin biopsies to predict progression and survival in a discovery cohort of 208 patients with CTCL (177 with MF) from a 15-year longitudinal observational clinical study. We compared these data to the results in an independent validation cohort of 101 CTCL patients (87 with MF). The tumor clone frequency (TCF) in lesional skin, measured by high-throughput sequencing of the TCRB gene, was an independent prognostic factor of both progression-free and overall survival in patients with CTCL and MF in particular. In early-stage patients, a TCF of >25% in the skin was a stronger predictor of progression than any other established prognostic factor (stage IB versus IA, presence of plaques, high blood lactate dehydrogenase concentration, large-cell transformation, or age). The TCF therefore may accurately predict disease progression in early-stage MF. Early identification of patients at high risk for progression could help identify candidates who may benefit from allogeneic hematopoietic stem cell transplantation before their disease becomes treatment-refractory.

INTRODUCTION

Cutaneous T cell lymphomas (CTCLs) are uncommon non-Hodgkin lymphomas of mature skin-tropic memory T cells. Mycosis fungoides (MF) is the most common and prevalent CTCL and typically presents as inflammatory patches and plaques on the skin. Diagnosis is often difficult and has relied on a combination of clinical, histopathological, and molecular findings (1). The average time from appearance of lesions to definitive diagnosis has been estimated to be 3 to 6 years (2). Recently, the advent of next-generation high-throughput DNA sequencing has revolutionized the diagnosis of MF. MF is nearly always a malignancy of CD4+ T cells with an αβ T cell receptor (TCR), encoded by the TCRA and TCRB genes (3). High-throughput sequencing of the TCRB gene can not only identify the unique T cell clone in MF but also precisely determine the tumor clone frequency (TCF) in the entire T cell infiltrate (3, 4).

A major challenge in the management of MF patients is the identification of early-stage patients at high risk for progression to advanced disease. More than 80% of early-stage patients will have an indolent lifelong course free of disease progression, regardless of treatment modality (5). As a result, early-stage patients are treated and maintained with conservative skin-directed therapies unless their disease worsens (6). However, a subset of patients develop highly aggressive, treatment-resistant disease that can be fatal. Although greater percent skin surface area involvement is associated with a somewhat higher risk of progression, most of early-stage MF patients have indolent courses (5). By contrast, advanced-stage patients (stage IIB or higher) have dismal prognoses, with life expectancies ranging from 1.5 to 4 years. Recently, allogeneic hematopoietic stem cell transplantation has emerged as a potentially life-saving intervention in advanced-stage CTCL patients (7). Outcomes from this procedure are somewhat better in patients with Sézary syndrome (SS; a leukemic form of CTCL) than with MF, but regardless, successful outcomes are observed only in patients who are in complete (or near complete) remission at the time of transplantation (8). Unfortunately, such significant remissions are typically impossible to achieve in advanced MF (9). Therefore, prospective identification of MF patients with early-stage disease who are at high risk for disease progression as potential candidates for allogeneic hematopoietic stem cell transplantation is a major unmet clinical need.

Much effort has been devoted to identifying early-stage patients at high risk for disease progression. Previous studies have identified clinical variables associated with decreased progression-free survival (PFS) (5, 10). A Cutaneous Lymphoma International Prognostic Index (CLIPI) has been developed and applied to patients with both early- and late-stage disease (11). Although useful in late-stage disease, when applied to independent cohorts of early-stage patients, this index has been of limited utility (12). Several studies have identified candidate biomarkers using transcriptional profiling that may improve the prognostic predictions in CTCL (1315), but these are cumbersome to use in clinical practice and none has been fully validated. Clinically useful and validated risk factors for progression in early-stage disease patients are still based on the physical exam. They include body surface area involvement (with CTCL disease stages T1/IA and T2/IB involving <10% and ≥10% body surface area, respectively) and the presence of skin plaques (subclass b) versus patches (subclass a) (table S1) (10). Although useful, these variables can be subjective, arbitrary, and imprecise; for example, stage T2/IB disease covers from 10 to 79% body surface area, and patients may have a mixture of patches and plaques in different proportions. An objective and quantitative biomarker that addresses likelihood of disease progression does not currently exist.

Recently, we showed that high-throughput sequencing of the TCR genes (TCRB and TCRG) provides a superior tool for the diagnosis of CTCL by precise identification of the malignant T cell clone (3). Because each T cell clone has a unique TCR complementarity-determining region 3 (CDR3) sequence (3), DNA sequencing allows the precise identification and absolute quantification of both malignant and benign T cell clones in CTCL (3, 4). Skin lesions of MF patients are infiltrated by large numbers of nonmalignant memory T cells, and it is often impossible to distinguish the malignant T cell clone from activated benign infiltrating T cells in early-stage lesions by histopathology (16). The high-throughput sequencing test greatly facilitates the diagnosis of early-stage disease (3), allows tracking of specific T cell clones over time and in different tissues (4, 17, 18), and detects residual disease after treatment with high sensitivity (19, 20). For the past three decades, the most commonly used diagnostic assay for clonality in CTCL patients uses polymerase chain reaction (PCR) amplification of the rearranged TCR gene, typically TCRG, followed by denaturing gradient gel electrophoresis and gel scanning or Biomed GeneScan analysis (21). These nonquantitative tests have false-negative rates of at least 25% and a false-positive rate of 15% in the setting of MF (21, 22) and are particularly unreliable in early-stage MF. Here, we asked whether high-throughput sequencing of TCRB in DNA extracted from lesional skin could predict clinical outcome in large cohorts of CTCL patients.

RESULTS

High-throughput TCRB sequencing in lesional skin of 309 patients with CTCL

We performed high-throughput sequencing of the TCRB gene in lesional skin of 309 patients with CTCLs in the DFCI-02016 longitudinal study at Dana-Farber between 2002 and 2016. The clinical characteristics of the 309 patients in the discovery cohort (n = 208) and validation cohort (n = 101) are detailed in tables S2 and S3, respectively. The distribution of the types of CTCL in both cohorts is shown in Fig. 1A and included primarily patients with MF and SS. The distribution of TCR Vβ family usage by the tumor clone is depicted in Fig. 1B. The most frequently used TCR Vβ family in CTCL patients was TCRBV20, which represented 13% of all T cell clones. Although our patients had MF, this observation is similar to published data in patients with SS (23). TRBV20 is associated with Staphylococcus aureus infection (24), which commonly colonizes the skin of CTCL patients and has been associated with superantigen-driven TCR stimulation in a subset of patients (24). We measured the TCF in each sample as follows: TCF = (v1/∑vn) × 100, where v1 is the number of reads of the most abundant TCRB sequence, and vn is the number of all rearranged TCRB sequence reads. This method does not take into account reactive T cells with two rearranged TCRB alleles, but these cells represent a minority of αβ T cells. Hence, the TCF is a conservative estimate of the TCF. Examples in two patients with stage 1B MF are depicted in Fig. 1C and fig. S1. Histopathological analyses demonstrated that a high TCF was not associated with higher absolute numbers of mononuclear cells in the skin infiltrate (Fig. 1D). There was no statistically significant difference in terms of TCF between patients with skin category T1 (<10% body surface area involved with patches and plaques), T2 (>10% body surface area involved with patches and plaques), and T4 (erythroderma) distribution. Only category T3 patients showed a small but statistically significant increase in TCF (P < 0.05). Thus, TCF did not increase as a function of skin category T value alone (Fig. 1E).

Fig. 1 High-throughput TCRB sequencing in 309 patients with CTCLs.

(A) Clinical diagnosis in 309 patients with CTCLs in the discovery and validation sets. Other: CD30+ lymphoproliferative disorder and CD8+ aggressive epidermotropic CTCL. Pre-Sézary refers to the evidence of blood abnormalities (B1; elevated absolute CD4+ T cell count or CD4+/CD8+ T cell ratio) that do not meet the criteria for stage B2 or SS (26). (B) TCRBV gene family usage by the malignant clone in 309 cases of primary CTCLs (C) Example of the measurement of the malignant clone frequency in the skin in two patients with stage IB MF. Three-dimensional (3D) histograms of the TCRB sequencing data in lesional skin in two patients with stage IB MF. On the upper panel, the 3D histogram shows the presence of a TCF of 8% (18,131 reads). This patient showed no evidence of disease progression after 4 years of follow-up. On the lower panel, the 3D histogram shows a TCF of 37% (264,252 reads; the y axis has been cut at 80,000). This patient died of disease progression after 28 months. (D) Hematoxylin and eosin sections of lesional skin biopsies in four patients with CTCL and various malignant clone frequencies and outcomes. Scale bars, 100 μm. Upper left: TCF (32% of the T cells) with progression after 2 years. Upper right: TCF (41% of the T cells) with progression after 2 months. Lower left: TCF (6% of the T cells) with no progression in 8 years. Lower right: TCF (6% of the T cells) with no progression in 9 years. (E) TCF according to the extent of body surface area involved in patients with MF. Medians are indicated by horizontal bars, and comparisons are carried out using Mann-Whitney U test; *P < 0.05 is considered significant. NS, not significant.

Immunostaining versus high-throughput TCRB sequencing

Counting T cells by immunostaining with antibodies to Vβ gene products has been used to identify clonal populations in the skin because all malignant clonal cells express the same Vβ gene product. Therefore, we asked whether immunostaining could substitute for high-throughput sequencing of the TCRB. However, antibodies are available for only about 50% of Vβ families. Moreover, immunostaining for Vβ is inherently imprecise in the identification and quantification of a specific T cell clone. In part, this is because a given TCRBV exon can rearrange and pair with one of 13 TCRBJ exons during intrathymic T cell maturation. In one patient (339) analyzed with high-throughput sequencing, 28.4% of skin T cells were TCRBV20, but only 39.8% of these TCRBV20 T cells shared the specific CDR3 sequence of the malignant clone (CSALGLSSYNEQFF) (fig. S2A). Thus, staining with the anti-Vβ20 antibody (fig. S2B) overestimated the true clonal frequency because it also stained benign infiltrating T cells expressing TCRBV20 (60.2% of T cells expressing TCRBV20; fig. S2C). In another patient (425), 68.4% of T cells were TCRBV05, but this population included 94% of the malignant clone (64.5% of total T cells) (TCRBV05-1/J01-02, CDR3 sequence CASSLGGTGGYTF; fig. S2, A to C). Here, antibody staining more accurately estimated the malignant clone but was still variable from histological section to section. These approaches appear to be fundamentally inferior at quantifying the malignant clone when compared to the highly quantitative metric of TCF.

Prognostic impact of clinical, histological, and molecular parameters

We first tested the association of the TCF in lesional skin, as measured by high-throughput sequencing of TCRB, with prognosis in all CTCL patients in our discovery cohort. A TCF of >25% in the skin was significantly associated with reduced PFS (P < 0.001) and overall survival (OS) (P < 0.001) in 208 patients with CTCL (Fig. 2A). This was confirmed in a validation cohort of 101 patients (P < 0.001) (Fig. 2B). The TCF in the skin was significantly associated with the PFS (P < 0.001) and OS (P < 0.001) in patients with MF, in which the disease primarily affects the skin (Fig. 2, C and D). By contrast, in patients with SS (in which there is considerable blood involvement), there was no significant association of the TCF in the skin with PFS or OS (Fig. 2E). This prompted us to restrict our subsequent analyses to patients with MF (n = 177, discovery cohort). We did not address the predictive value of TCF in the blood of SS patients in this study because we were focused on the utility of skin biopsy alone.

Fig. 2 The TCF in the skin as predictor of PFS and OS in patients with CTCLs.

(A) Kaplan-Meier estimates of PFS (left) and OS (right) in 208 patients with CTCLs in the discovery set, according to the TCF in the skin (≤25% versus >25% of the total T cells in the skin). (B) Kaplan-Meier estimates of PFS in 101 patients with CTCLs in the validation set, according to the TCF in the skin (≤25% versus >25% of the total T cells in the skin). (C) Kaplan-Meier estimates of PFS (left) and OS (right) in 177 patients with MF in the discovery set, according to the TCF in the skin (≤25% versus >25% of the total T cells in the skin). (D) Kaplan-Meier estimates of PFS in 87 patients with MF in the validation set, according to the TCF in the skin (≤25% versus >25% of the total T cells in the skin). P values in (A) to (D) are estimated by Cox univariable analysis. (E) Kaplan-Meier estimates of PFS (left) and OS (right) in 22 patients with SS in the discovery set, according to the TCF in the skin (≤25% versus >25% of the total T cells in the skin).

Gender, age, blood lactate dehydrogenase (LDH) concentration, folliculotropism, large-cell transformation, and the presence of a clone in the skin detected by PCR have all been associated with disease progression in MF (5). We therefore compared these variables to the TCF in our discovery cohort for their association with PFS and OS. Age (>60 years) (P < 0.01), elevated LDH (P < 0.01), the existence of large-cell transformation (P < 0.001), and the TCF of >25% (P < 0.001) were significantly associated with PFS and OS (P < 0.001, P = 0.001, P = 0.001, P < 0.001, respectively) in univariable analysis (Table 1). In a multivariable analysis that included age, advanced tumor stage, LDH concentration, large-cell transformation, and the type of treatment received as confounding factors, the TCF was still significantly associated with PFS (P < 0.001) and OS (P < 0.001). A TCF of 25% was found to be the best cutoff, as determined by the concordance index (25) in univariable analysis on PFS and OS. There was a relatively continuous relationship between the TCF threshold and the hazard ratios (HRs) for OS and PFS until a TCF of 25% where a plateau was reached (fig. S3).

Table 1 Uni- and multivariable analysis on PFS and OS in 177 patients with MF in the discovery set.

Treatments used before first evidence of progression, death, or censoring. Phototherapy includes PUVA (psoralen + ultraviolet A) and UVB therapy. Radiation therapy includes electron-beam therapy and brachytherapy. Systemic treatments include interferon-α, oral bexarotene, folate inhibitors, systemic histone deacetylase inhibitors, and monoclonal antibodies. The multivariable model was stratified on LDH levels and the use of systemic treatments. CI, confidence interval.

View this table:

Prognosis in early-stage MF

We hypothesized that the predictive value of the TCF in the skin might be more useful in patients with early-stage MF (skin-limited disease), where outcome is uncertain, than in patients with advanced-stage disease, where life expectancy is invariably reduced. There was a significant interaction between the TCF and the disease stage (early-stage versus advanced-stage, P < 0.01). This prompted us to study the prognostic value of the TCF in the skin in patients with early-stage MF and compare it to existing prognostic factors in these patients. Most patients with MF present with early-stage disease, and up to 20% will experience disease progression and/or death within 10 years (5). In early-stage patients with skin-limited disease, the body surface area involved in the disease is considered the primary prognostic factor, with T1/IA involving <10% and T2/IB involving ≥10% of the body surface area (table S1) (5, 26). Variables significantly and independently associated with prognosis in the entire cohort (Table 1) were studied in a subcohort of early-stage patients. Univariable analysis on 141 early-stage MF patients in the discovery set revealed that a TCF of >25% (P < 0.001), disease stage (T2/IB versus T1/IA) (P < 0.01), age (>60 years) (P < 0.05), and the presence of plaques (P < 0.05) were each significantly associated with PFS, but the HR for TCF was the highest (Table 2). To compare the prognostic value of the TCF to the reference prognostic index in patients with early-stage CTCL, the CLIPI was calculated in these patients, as described in (11). An intermediate or high CLIPI score was associated with a lower PFS in the discovery set, but again, the HR was lower than that of the TCF (Table 2 and fig. S4).

Table 2 Prognostic factors of PFS in 141 patients with early-stage disease from the training set and 69 patients in the validation set.
View this table:

A TCF of >25% showed the highest HR for PFS in both the discovery set (HR, 4.9; 95% CI, 2.5 to 9.7; P < 0.001) and the validation set (HR, 10; 95% CI, 3.4 to 31; P < 0.001) (Fig. 3, A and B). By contrast, the HR of stage (T2/IB versus T1/IA) was lower in both the discovery set (HR, 2.5; 95% CI, 1.3 to 4.9; P < 0.01) and the validation set (HR, 6.4; 95% CI, 1.5 to 28; P = 0.01). In the discovery set, 89% (95% CI, 76 to 95) of stage T2/IB patients with a TCF <25% were alive without disease progression 4 years later versus 30% (95% CI, 7 to 58) of stage T2/IB patients with a TCF >25% (Fig. 3C). In the validation set, 85% (95% CI, 50 to 96) of stage T2/IB patients with a TCF <25% were alive and progression-free 4 years later versus 19% (95% CI, 5 to 40) of patients with a TCF >25% (Fig. 3D). The TCF in the skin was also significantly associated with OS (P < 0.01) (Fig. 3C).

Fig. 3 The TCF in the skin as predictor of PFS and OS in patients with early-stage MF.

(A) Kaplan-Meier estimates of PFS (left) and OS (right) in 141 patients with early-stage (IA to IIA) MF in the discovery set, according to the TCF (<25% versus >25% of the total T cells in the skin). (B) Kaplan-Meier estimates of PFS in 69 patients with early-stage (IA to IIA) MF in the validation set, according to the TCF (≤25% versus >25% of the total T cells in the skin). (C) Kaplan-Meier estimates of PFS (left) and OS (right) in 70 patients with stage IB MF in the discovery set, according to the TCF (≤25% versus >25% of the total T cells in the skin; top) or to the presence of plaques (bottom). (D) Kaplan-Meier estimates of PFS in 42 patients with stage IB MF in the validation set, according to the TCF (≤25% versus >25% of the total T cells in the skin; top) or to the presence of plaques (bottom). P values in (A) to (D) are estimated by Cox univariable analysis. (E) Dot plot and linear regression of the time to progression/death according to the TCF in the skin in stage IB patients from the discovery and validation sets who experienced disease progression during the follow-up. Pearson’s correlation coefficient and P value are indicated. (F) Receiver operating characteristic curve of the TCF in the skin (>25%) in patients with stage IB MF in the discovery and validation sets for 5-year progression or death. Progressors are patients who progressed or died within 5 years after the test. Nonprogressors are patients with at least 5 years of follow-up and no event of death or progression in 5 years. The sensitivity is defined as the percentage of patients with a malignant clone (>25% of T cells in the skin) among progressors. The specificity is defined as the percentage of patients with a malignant clone (≤25% of T cells in the skin) among nonprogressors.

There was a significantly higher TCF in patients with plaques compared to patches (P < 0.001). However, a number of patients with patches had a high TCF in the skin, and conversely, many patients with plaques had a low TCF (fig. S5). To directly compare the prognostic value of TCF to both skin stage (T1/IA versus T2/IB) and the presence of plaques (b) versus patches (a), we assessed these variables on the discovery and validation sets. Figure S6 compares PFS and OS of T1a, T1b, T2a, and T2b patients and confirms that both skin stage (T2/IB versus T1/IA) and the presence of plaques are associated with decreased PFS (P < 0.01 and P < 0.05 for skin stage and plaques, respectively) and that the skin stage is associated with decreased OS (P < 0.01) in Cox univariable analysis. However, when PFS and OS in stage IB/T2 patients were assessed according to the presence or absence of plaques (IB/T2a versus IB/T2b) or a TCF of >25% (Fig. 3, C and D), the latter was far more predictive. For the TCF of >25%, stage IB/T2 patients had decreased PFS (HR, 13; 95% CI, 5 to 36; P < 0.001 in the discovery set; HR, 11; 95% CI, 2.5 to 48; P = 0.001 in the validation set) and OS (HR, 9.0; 95% CI, 3.0 to 27; P < 0.001) (Fig. 3, C and D). For T2b versus T2a, the HR for PFS was 2.2 (95% CI, 0.9 to 5.3; P = 0.08) in the discovery set and 1.7 (95% CI, 0.6 to 4.8; P = 0.3) in the validation set, and the HR for OS was 2.0 (95% CI, 0.7 to 6.5; P = 0.2) (Fig. 3, C and D). The TCF was significantly associated with PFS (HR, 5.8; 95% CI, 1.8 to 19; P = 0.004) in a multivariable model that included the age, stage T2 versus T1, and the presence of plaques as covariates (table S4). This was confirmed on the independent validation cohort (HR, 13.6; 95% CI, 1.2 to 154; P = 0.03). Therefore, in stage IB/T2 patients, a TCF of >25% was highly predictive of PFS and OS and was far more predictive than the presence of plaques versus patches. In stage IB patients who experienced progression or death during the follow-up, there was an inverse correlation between the TCF and the time to progression or death (ρ = −0.6, P < 0.001) (Fig. 3E).

A TCF of >25% in the skin was associated with a positive predictive value of 92% for 5-year disease progression or death and a negative predictive value of 83% (Fig. 3F). As previously shown (5), stage IA/T1 patients (who have limited skin involvement with <10% of the body surface area involved) had an excellent prognosis regardless of the TCF (fig. S7). These data indicate that the frequency of the malignant T cell clone in the skin (TCF) is the single best predictive test for identifying patients at risk for disease progression, particularly in stage IB patients, who appear to be the only early-stage patients who progress. The variability of the TCF between different lesions of the same type (for example, patches or plaques) in the same patient was low, as depicted in fig. S8. There was no significant difference between early-stage patients with a TCF of >25% and ≤25% in terms of treatments received before sequencing. Patients with a TCF of >25% had a poor PFS and OS with no significant difference between treatment-naïve and pretreated patients. The TCF was significantly associated with PFS (HR, 4; 95% CI, 1.3 to 12; P = 0.01) and OS (HR, 8.9; 95% CI, 2 to 39; P = 0.004) in treatment-naïve early-stage patients (fig. S9).

T cell microenvironment

We speculated that the poor prognosis associated with a higher TCF in the skin might be linked to a defective antitumor immune response, an intrinsic aggressiveness of the malignant cells themselves, or a combination of these variables. To investigate the causal mechanism, we assessed the immune microenvironment in patients with a high TCF in the skin versus patients with a low TCF in the skin. Because the TCF accounts for the number and diversity of nonmalignant T cells present in the lesional skin biopsy, differences in the TCF might simply represent different numbers of reactive, nonmalignant T cells in the setting of similar numbers of clonal T cells.

An increased number of reactive CD8+ T cells in the skin of patients with CD4+ MF has been previously associated with improved prognosis (27, 28). We performed CD8 and granzyme B staining in the skin of early-stage MF patients with a high TCF (>30% T cells) and a low TCF in the skin (<10%) (Fig. 4A). TCFs of 10 and 30% were chosen as cutoffs because they were close to the 25th and 75th percentiles in this population. There was no significant difference in the percentage of CD8+ T cells in the skin between these two groups. The percentage of granzyme B–positive cells was not significantly lower in patients with a high TCF (Fig. 4B). Patients with a high TCF in the skin had a more clonal reactive T cell infiltrate (Fig. 4C), a feature that has been suggested to represent the capacity to mount an antitumor immune response in the skin (29). Thus, a defective antitumor immune response alone, by these criteria, does not seem to be the primary mechanism of the progression in patients with a high TCF.

Fig. 4 Samples with a high TCF are not associated with a decreased antitumor immune response.

(A) Example of CD8+ and granzyme immunostaining in lesional skin in two lesional CTCL skin biopsies. DAPI, 4′,6-diamidino-2-phenylindole. (B) CD8+ T cell percentage (left) and granzyme B–positive cell percentage in lesional skin of CTCL patients with a low TCF (<10% T cells) and high TCF (>30% T cells) (*P < 0.05 and **P < 0.01, Mann-Whitney U test). (C) Reactive T cell clonality (left) and entropy (right) in lesional skin of CTCL patients with a low TCF (<10% T cells) and high TCF (>30% T cells) (*P < 0.05, Mann-Whitney U test).

Gene expression profiling and exome sequencing

We obtained gene expression data on 78 potential biomarkers in CTCL that were obtained in lesional skin biopsies from 157 patients with early- and advanced-stage MF and SS. These potential biomarkers were selected on the basis of previous studies showing their overexpression in CTCL compared to normal skin (1315) or their copy number variations in CTCL (23, 27, 28, 30). The unsupervised analysis of the data set revealed that patient transcriptomes clustered in three groups according to gene expression (Fig. 5A), in accordance with previously published works (1315). Patients in cluster 1 overexpressed numerous T cell–specific genes, such as cell surface markers (CD4, CCR4, CCR7, CD28, CD52, and PDCD1), genes in the interleukin-21 (IL-21)/Janus kinase (JAK)/signal transducer and activator of transcription (STAT) pathway (IL21, IL2RG, and JAK3), and genes in the TCR signaling pathway (ITK, LCK, PRKCQ, SH2D1A, FYB, LAT, PTPRCAP, RAC2, GIMAP4, T3JAM, CARD11, SIT1, PIK3CD, VAV1, and LEF1). We next asked whether this gene expression pattern simply reflected differences in T cell abundance between cluster 1 and the two other clusters. Although there were no statistically significant differences in the absolute abundance of total T cells in any of the clusters studied (Fig. 5B), the abundance of the malignant clone (relative to total lesional T cells) was significantly higher in patients in cluster 1 (poor prognosis) compared to the two other clusters (P < 0.05 compared to cluster 3 and P < 0.01 compared to cluster 2, Mann-Whitney U test with Bonferroni correction) (Fig. 5C). There was a significantly lower PFS in patients in cluster 1 versus the two other clusters (P < 0.05 compared to cluster 3 and P < 0.001 compared to cluster 2, log-rank test with Bonferroni correction) (Fig. 5D). Thus, we identified a cluster of patients (cluster 1) with a distinct gene expression profile, high TCF in the skin, and poor prognosis. The overexpression of genes in the JAK-STAT and TCR signaling pathways in patients with a high TCF is consistent with the role of these pathways in T cell proliferation and survival.

Fig. 5 A high TCF in the skin is associated with a distinct gene expression profile and a higher number of somatic mutations.

(A) Unsupervised analysis by hierarchical clustering (complete linkage) according to the expression of 78 genes in 157 patients reveals three different clusters of patients. Intensity expression values in the heat map are expressed as log2 fold changes compared to the average expression of each gene in the whole study group. The TCF in each sample is represented by a color scale at the bottom of the heat map. (B) Dot plots of the T cells (frequency of nucleated cells) in patients in clusters 1 to 3. Medians were compared by Mann-Whitney U test with Bonferroni adjustment for multiple testing (*P < 0.05 and **P < 0.01). (C) Dot plots of the TCF in patients in clusters 1 to 3. Means were compared by Mann-Whitney U test with Bonferroni adjustment for multiple testing (*P < 0.05 and **P < 0.01). (D) Kaplan-Meier estimates of PFS in 157 patients with CTCLs in the training group, according to the gene expression clustering. Log-rank test with Bonferroni adjustment for multiple testing (*P < 0.05, **P < 0.01, and ***P < 0.001). (E) Whole-exome sequencing data of microdissected skin T cells in patients with MF. Left: Number of somatic mutations according to the clinical stage. *P < 0.05, Mann-Whitney U test. Right: Number of somatic mutations according to the malignant clone frequency in the skin. P < 0.05 by Spearman correlation is considered significant.

Whole-exome sequencing of tumors has yielded valuable data in a variety of cancers but has been difficult to perform in patches or plaques of MF because of the paucity of tumor cells relative to total nucleated cells. We thus conducted whole-exome sequencing on microdissected skin T cells in 19 patients with skin-limited MF using peripheral blood mononuclear cells as a comparator. The mean target coverage was 70× in tumor samples and 103× in peripheral blood mononuclear cells. The number of somatic mutations was significantly correlated with the TCF in the skin (ρ = 0.5, P = 0.04) (Fig. 5E). There were a higher number of somatic mutations in patients in stage IIB compared to patients in stage IB and IA, but all demonstrated abundant somatic mutations. This indicates that the poor prognosis associated with a high TCF in the skin is accompanied by a high frequency of genetic abnormalities of the malignant cells.

DISCUSSION

Here, we showed that an increased frequency of the malignant T cell clone in the skin was strongly correlated with reduced PFS and OS in patients with CTCL, particularly in patients with early-stage MF with a T2 distribution. Histological examination of patches and plaques for the density of the T cell infiltrate could not distinguish lesions with high TCF versus low TCF. TCF was the single best independent predictor of PFS and OS in MF, particularly in early-stage MF. Moreover, it is readily obtained from sequencing a small skin biopsy. Before this study, assessment of body surface involvement (T1 versus T2) has been the best means of predicting which patients might progress. Although our data agree with the well-accepted observation that stage IB patients (T2) are more likely to progress than stage IA (T1) patients, the TCF outperformed the predictive value of T category. Similarly, the presence of skin plaques (versus patches) and the currently used CLIPI criteria were less discriminative and predictive than the TCF in our early-stage patients. Together, our findings suggest that a TCF of >25% in MF skin lesions is the most sensitive and specific method available to identify early-stage patients at the highest risk for disease progression. For patients with T2/stage IB disease, the positive predictive value is 92%, and the negative predictive value is 83% for 5-year PFS.

Risk stratification is one of the goals of precision oncology, and there is great interest in biomarkers that predict aggressive disease in malignancies in which a majority of patients have indolent disease, whereas a smaller subset has aggressive disease. Identifying patients at risk for disease progression is particularly important in CTCL, a disease in which two patients with similar physical exams and histopathological morphology can have markedly different outcomes. High-throughput sequencing of the TCRB gene provides a quantitative measurement of the malignant T cell clonal burden in CTCL lesions. Moreover, it is straightforward and readily accomplished using available platforms.

It is possible that differences in the TCF might simply represent different numbers of reactive, nonmalignant T cells in the setting of similar numbers of clonal T cells. However, the total number of T cells alone was not able to accurately discriminate patients at high risk of progression. An increased number of reactive CD8+ T cells in the skin of patients with CD4+ MF has been previously associated with improved prognosis (31, 32). However, our data indicate that a defective antitumor immune response does not appear to be the primary mechanism associated with progression in patients with a high TCF in the skin. The subset of patients with a high TCF displayed a specific gene expression profile in lesional biopsies, and their malignant T cells harbored a higher number of somatic mutations. These alterations were already detectable in stage IB patients with a TCF of >25% in the skin, which is consistent with the poor prognosis associated with these features.

The limitations of our study include the fact that this is a retrospective analysis of patients from a single cancer center. Patients’ samples were collected at the time of presentation to the Dana-Farber Cancer Institute (DFCI) Cutaneous Lymphoma Clinic. Many patients were treatment-naïve, but others were referred from outside facilities for disease management, often after relapse or progression during treatment. This precluded us from taking samples exclusively upon diagnosis and before treatment. However, treatment appeared to have no effect on the predictive nature of TCF. In patients with a TCF of >25%, there was no difference in terms of PFS or OS between patients who had received treatments before inclusion and treatment-naïve patients (fig. S9). The usually indolent nature of MF and the long follow-up period required mean that prospective studies will take many years; however, such studies are essential to validate our findings.

The prognostic value of a TCF of >25% in the skin for PFS and OS may help identify the subset of patients who may ultimately benefit from allogeneic stem cell transplant earlier in their disease course. Because the tumor cells are restricted to the skin in MF, their overall tumor burden is a product of the TCF and the body surface area involved. It is therefore not surprising that a high TCF is associated with a poor prognosis in these patients. Overexpression of genes involved in TCR signaling was found in MF patients with a high TCF, and in peripheral T cell lymphomas, this has led to therapeutic strategies that target TCR signaling (33). Our gene expression data suggest that therapeutic inhibition of TCR signaling in CTCL may be a useful strategy, although other pathways might be involved in CTCL cell survival and proliferation. Finally, the role of immunomodulatory therapies in this subset of high-risk patients should be assessed. A previously published study showed efficacy of topical resiquimod, a Toll-like receptor agonist, including in patients with a high pretreatment TCF in the skin (19). In summary, our data suggest that a TCF of >25%, as determined by high-throughput DNA sequencing of the TCRB gene, is a strong predictor of disease progression and decreased survival in patients with MF limited to the skin.

MATERIALS AND METHODS

Study design

This is an experimental laboratory study performed on human tissue samples. All studies were performed in accordance with the Declaration of Helsinki. Lesional skin from patients with CTCL was obtained from patients seen at the Dana-Farber/Brigham and Women’s Cancer Center Cutaneous Lymphoma Program and included in the DFCI-02016 observational cohort, after informed consent. Eligibility criteria included a confirmed diagnosis of CTCL after review of the clinical, molecular, and histological data, as well as adequate remaining research or clinical specimens for high-throughput sequencing. The CTCL patients studied met the WHO-EORTC (World Health Organization–European Organization for Research and Treatment of Cancer) criteria for SS or MF (22). All tissues were collected with previous approval from the Partners and Dana-Farber Institutional Review Boards. All samples with enough available material were analyzed by high-throughput sequencing of the TCRB gene, which were masked to clinical outcomes. Staging and disease progression were evaluated according to the International Society for Cutaneous Lymphomas (ISCL)/EORTC criteria (10, 26). Analyses of high-throughput sequencing data were carried out in an investigator-blinded fashion. Immunostaining studies were performed using in vitro assays without blinding or randomization. Study components were not predefined.

Patients

The primary discovery cohort comprised 208 patients with CTCL seen at the Dana-Farber Cancer Institute’s Cutaneous Lymphoma Clinic from 2002 to 2016 (table S2). This discovery set included 177 patients with MF with a median follow-up of 8 years. Samples were typically collected at the time of diagnosis or at the time of referral to the DFCI Cutaneous Lymphoma Clinic for management of established disease. The independent validation set included 101 distinct CTCL patients recently included in the same study, including 87 patients with MF (table S3). The data were collected on 23 December 2016.

Nucleic acid extraction

DNA and RNA were extracted from four 20-μm-thick formalin-fixed, paraffin-embedded (FFPE) tissue scrolls from a lesional skin biopsy using the AllPrep DNA/RNA FFPE Isolation kit (Qiagen), as per the manufacturer’s instructions. DNA and RNA amounts were measured using a BioDrop spectrophotometer (Denville Scientific Inc.). For fresh frozen samples from the validation set, DNA was isolated from 30 cryosections with a thickness of 10 μm. DNA extraction was carried out using the QIAamp DNA Mini kit (Qiagen), as per the manufacturer’s instructions, with overnight tissue digestion.

High-throughput sequencing of the TCRB gene

Immunosequencing. For each sample, DNA was extracted from skin biopsies. We then shipped it on dry ice to Adaptive Biotechnologies. TCRB CDR3 regions were amplified and sequenced using ImmunoSEQ (Adaptive Biotechnologies). The ImmunoSEQ platform is available as a kit or service (www.adaptivebiotech.com/immunoseq). All TCRB characterization was performed by Adaptive Biotechnologies using the ImmunoSEQ TCRB “survey level” human assay (4, 34), which was previously described in detail (3). Bias-controlled V and J gene primers were used to amplify rearranged V(D)J segments for high-throughput sequencing at ~20× coverage. After correcting sequencing errors via a clustering algorithm, CDR3 segments were annotated according to the international ImMunoGeneTics collaboration, identifying which V, D, and J genes contributed to each rearrangement (35, 36).

Controlling bias in a multiplex PCR. Because accurate quantification of lymphoblast clones for minimal residual disease detection is critical, an approach was developed to ensure minimal bias in multiplex PCR. Briefly, each potential V(D)J rearrangement of the TCRB locus contains 1 of 13 J segments, 1 of 2 D segments, and 1 of 52 V segments, many of which have disparate nucleotide sequences. To amplify all possible V(D)J combinations, a single-tube, multiplex PCR assay with 45 V forward and 13 J reverse primers was used. To remove potential PCR bias, every possible V-J pair was chemically synthesized as a template with specific barcodes (34). These templates were engineered to be recognizable as nonbiologic and have universal 3′ and 5′ ends to permit amplification with universal primers and subsequent quantification by high-throughput sequencing. This synthetic immune system was then used to calibrate the multiplex PCR assay. Iteratively, the multiplex pool of templates was amplified and sequenced with TCRB V/J-specific primers, and the primer concentrations were adjusted to rebalance PCR amplification. Once the multiplex primer mixture amplified each V and J template nearly equivalently, residual bias was removed computationally.

PCR template abundance estimation. To estimate the average read coverage per input template in the multiplex PCR and sequencing approach, a set of about 850 unique types of synthetic TCR analog comprising each combination of Vβ and Jβ gene segments were used. These molecules were included in each PCR at very low concentrations so that most unique types of synthetic template were not observed in the sequencing output. Using the known concentration of the synthetic template pool, the relationship between the number of observed unique synthetic molecules and the total number of synthetic molecules added to the reaction was simulated (this is very nearly one-to-one at the low concentrations that were used). These molecules then allowed calculation for each PCR of the mean number of sequencing reads obtained per molecule of PCR template (the amplification factor) and, thus, estimation of the number of T cells in the input material bearing each unique TCR rearrangement.

Clonal detection. The putative malignant clone was defined by sequence abundance. A clone can have either one or two rearranged TCR alleles. For most of the clones, both TCRG alleles are rearranged, and for TCRB, a minority has both alleles rearranged. For consistency, a clone’s abundance was defined by summing the abundance of the most frequent single allele for TCRB. The putative malignant clone was defined by relative abundance of its unique CDR3 sequence (3). The percent of T cells consisting of the malignant clone was determined by dividing the abundance of the malignant clone (number of reads) by the total number of T cell reads. This method does not take into account reactive T cells with two rearranged TCRB alleles, but these cells represent a minority of αβ T cells.

Reactive T cell diversity and clonality measurements. The diversity of the reactive T cell clones was studied using the Shannon’s index. Shannon’s entropy quantifies the uncertainty in predicting the sequence identity of a random sequence from a data set. The Shannon’s index of the reactive clones (H) was calculated according to the following formula: H Embedded Image, where i represents each individual reactive clone, and f represents the frequency of this rearrangement among all productive rearrangements in the sample, excluding the malignant clone. To allow for comparisons between samples differing in the total number of reads, entropy was normalized by division of log2 of the number of unique productive sequences. Clonality is the reciprocal of normalized Shannon’s entropy (clonality = 1 − normalized entropy) with values ranging from 0 (most diverse) to 1 (least diverse).

Cryosection immunostaining and cell counting

CTCL skin samples were coimmunostained for anti-Vβ2 (clone MPB2D5, Beckman-Coulter) or anti-Vβ5.1 (clone IMMU 157, Beckman Coulter) conjugated to R-phycoerythrin with anti-CD3 conjugated to Alexa Fluor 647 (clone UCHT1, BioLegend) with three 5-min wash steps in tris-buffered saline–saponin before mounting. Single-color controls confirmed specificity of staining and no bleed through into the other channel. The samples were analyzed using an Olympus BX43 microscope with the objective lens of 10×/0.40, 20×/0.75, and 40×/0.95 Olympus UPlanFL (Olympus). Images were acquired with the Mantra Quantitative Pathology Imaging System and analyzed using inForm software (PerkinElmer) and the manual counting feature from Adobe Photoshop CS5 (Adobe). We analyzed 10× images of nonoverlapping fields.

Transcriptional analyses

At least 140 ng of mRNA per sample was analyzed by NanoString gene expression profiling using a custom CodeSet including 78 probes directed against potential biomarkers identified in previous gene expression studies by our group (1315) or in exome sequencing studies by others (23, 27, 28, 30) and three housekeeping genes. The NanoString technology uses molecular barcode and single-molecule imaging for the direct hybridization and detection of hundreds of unique transcripts in a single reaction. Each color-coded barcode is attached to a single target-specific probe corresponding to an analyte of interest. Combined together with invariant controls, the probes form a multiplexed CodeSet. The samples are run on the nCounter platform. Gene expression data were background-subtracted and normalized to positive controls and housekeeping genes using the NanoString nSolver software (37). Gene expression values were expressed as log2 fold changes of the average gene expression of the considered gene in the whole study group. Gene expression assays were performed blinded to the patient’s outcome.

Exome sequencing of microdissected skin T cells

Sample preparation and expression microdissection. The total lesional skin biopsies of 19 patients with skin-limited MF were embedded in optimal cutting temperature compound and stored frozen at −80°C. Slides (6 μm) were then sectioned using a microtome cryostat and stained for CD3 by immunohistochemistry. Briefly, slides were blocked with bovine serum albumin and incubated with rabbit recombinant anti-human CD3 antibody (Sp7, Abcam), followed by secondary antibody coupled to horseradish peroxidase (Envision+ Dual Link, DAKO) and revelation with diaminobenzidine (Vector Laboratories). Slides were dehydrated in alcohol and xylene. A membrane was placed on the tissue, and a flashlamp was applied. The flashlamp is an intense pulsed light that emits a bright range of wavelengths from ultraviolet to visible light and infrared, but ultraviolet light is filtered out and does not reach the tissue. The light excites and heats the stained cells that transfer to the membrane. The membrane was then placed in lysis buffer, and DNA was extracted using a QIAamp DNA Micro kit (Qiagen). The DNA quantity and integrity were measured by using a Bioanalyzer. A matched blood sample from the same patient, without blood involvement as confirmed by high-throughput sequencing of the TCRβ gene, was used as a germline control.

Library preparation and sequencing

This study was carried out in collaboration with the Center for Cancer Genome Discovery at Dana-Farber. Before library construction, DNA was fragmented (Covaris sonication) to 250 bp (base pairs) and further purified using Agencourt AMPure XP beads. Size-selected DNA was then ligated to specific adaptors during manual library construction [Modified (low input) KAPA Library Prep]. Each library was made with sample-specific barcodes, quantified using the MiSeq, and libraries were pooled at equal mass (1 × 2-plex) to a total of 750 ng for Exome v5 enrichment using the Agilent SureSelect hybrid capture kit. The capture was then sequenced on HiSeq 2500 and 3000.

Variant analysis

Mutation analysis for single-nucleotide variants was performed using MuTect v1.1.4 (38) and annotated by Variant Effect Predictor. We used the Somatic Indel Detector tool that is part of the Genome Analysis Toolkit for indel calling. MuTect was run in paired mode, pairing the tumor sample to the matched normal.

Quality control

Eighty percent of the targets were covered at least 20×. Fingerprinting analysis was performed using 44 polymorphic loci to identify whether the aggregation pairing strategy was performed appropriately. Picard Tools Genotype Concordance was used to calculate the concordance that a given test sample matches the sample being considered. This was performed on all pairwise combinations of samples in the cohort. The output of the pairwise comparisons was then mapped to a concordance matrix, where concordance values above four SDs of the median concordance value for the cohort indicated a high likelihood that the samples match.

Statistical analyses

Patient clinical information was collected at the reference date of 23 December 2016. PFS and OS were estimated by the Kaplan-Meier method. PFS was defined as the time between sampling and death from any cause or progression of the lymphoma disease or the last disease evaluation time where no disease progression was observed. OS was defined as the time between sampling and death from any cause (OS) or censoring at the last follow-up. Age, gender, disease stage, serum LDH concentrations, the existence of folliculotropism or large-cell transformation, the presence of a clonal pattern in the skin as assessed by PCR of the TCRγ gene TCRG, the malignant clone frequency as assessed by high-throughput sequencing of the TCRB gene, and the CLIPI were assessed for their association with PFS and OS in univariable Cox regression analysis. The CLIPI was calculated on the basis of the presence of the following factors: age (>60 years), male sex, plaques, folliculotropism, and clinical adenopathy N1/Nx; low risk = 0 to 1, intermediate risk = 2, and high risk = 3 to 5 prognostic factors (11). Missing values for LDH were imputed in five patients (mean imputation based on disease stage), and all cases were used in the final analysis. For PCR, the analysis was carried out on complete cases (n = 114). A stepwise selection process was applied, with all variables significant (P < 0.05) in univariable analysis retained in the initial multivariable model, followed by backward elimination. Interactions between the clone frequency and other variables were tested. The proportionality assumption was tested for each variable in the final model, and the model was stratified on variables that violated the proportionality assumption. The model was then tested on an independent cohort. The cutoff value for the malignant clone frequency was selected to maintain the concordance index and Akaike’s information criterion of the univariable model using the corresponding continuous variable. Sixty years was chosen as the cutoff for age because it was the most frequently selected cutoff in previously published studies in the field. OS was not tested in the validation set due to recent sampling of the patients leading to an insufficient number of events. Comparisons of the T cell percentages and TCF between the three clusters were performed using the Mann-Whitney U test, followed by Bonferroni correction for multiple testing, with P < 0.05 considered significant. Heat maps and hierarchical clustering (complete linkage) were performed with Genesis software (Institute for Genomics and Bioinformatics, Graz Institute of Technology, Graz, Austria; available at http://genome.tugraz.at/genesisclient/genesisclient_description.shtml). Medians and interquartile ranges are indicated on dot plots. Statistical analyses were performed with R 3.1.1 and GraphPad Prism.

SUPPLEMENTARY MATERIALS

www.sciencetranslationalmedicine.org/cgi/content/full/10/440/eaar5894/DC1

Fig. S1. Clinical pictures of two patients with stage IB MF.

Fig. S2. TCR Vβ high-throughput sequencing allows specific quantification of the frequency of the malignant T cell clone within a Vβ gene family.

Fig. S3. Continuous relationship between the TCF and the HRs for PFS and OS.

Fig. S4. Prognostic value of the CLIPI in early-stage MF.

Fig. S5. TCF in patches versus plaques.

Fig. S6. Prognosis in early-stage patients according to body surface area involved and the presence of plaques.

Fig. S7. Prognosis in stage IA patients.

Fig. S8. Reproducibility of the TCF as measured by high-throughput sequencing of the TCRβ gene in different lesions in the same patient.

Fig. S9. PFS and OS in pretreated and treatment-naïve early-stage MF patients.

Table S1. ISCL/EORTC classification and staging of MF and SS.

Table S2. Clinical characteristics of 208 patients with CTCL in the discovery set.

Table S3. Clinical characteristics of 101 patients with CTCL in the validation set.

Table S4. Multivariable analysis on PFS in early-stage patients.

REFERENCES AND NOTES

Acknowledgments: We thank I. Litvinov for the help with data collection and the statistical reviewers from Harvard Catalyst for the helpful statistical advice and reviewing of the paper, as well as A. Thorner, A. Nag and B. Wollison from the Dana-Farber Center for Cancer Genome Discovery for contributing the exome sequencing data. We also thank F. Kuo and the Pathology Specimen Locator Core of the Dana-Farber Harvard Cancer Center. Funding: This study was supported by charitable contributions from E. P. Lawrence and from the Lubin Family Foundation. Support was also obtained from NIH Specialized Program of Research Excellence grant P50 CA9368305 (to T.S.K.), NIH R01CA203721 (to R.A.C. and T.S.K.), NIH T32 AR007098 (to T.S.K.), NIH R01 AR063962 (to R.A.C.), and NIH P30 AR069625 (to R.A.C.). Support also came from grants from the Société Française de Dermatologie, Collége des Enseignants de Dermatologie de France, Association pour la Recherche contre le Cancer, Fondation Rene Touraine, and the Philippe Foundation (to A.d.M.) and a Curing Cutaneous Lymphoma by Advancing Research, Innovation and Offering New Solutions grant from the Cutaneous Lymphoma Foundation (to J.T.O). This work was conducted with support from Harvard Catalyst at the Harvard Clinical and Translational Science Center (National Center for Research Resources and the National Center for Advancing Translational Sciences, NIH award UL1 TR001102 to J.T.O) and financial contributions from Harvard University and its affiliated academic health care centers. Author contributions: A.d.M., J.T.O., R.A.C., and T.S.K. designed the study, performed the data analysis and interpretation, and wrote the manuscript. A.d.M., E.L.L., J.E.T., J.T.O., and C.P.E. did the laboratory experimentation and analysis. H.R. and I.R.K. contributed to the data analysis of high-throughput sequencing. S.S.G., J.T.O., S.J.D., M.T., D.C.F., P.M.D., E.L.L., N.R.L., and T.S.K. acquired the clinical samples and patient data and provided the clinical interpretation. A.d.M. and J.T.O. prepared the figures. All authors approved the final manuscript. Competing interests: H.R. and I.R.K. are employed by and own equity in Adaptive Biotechnologies. T.S.K., R.A.C., J.T.O., and A.d.M. are inventors on patent application US 62/653,854 submitted by The Brigham and Women’s Hospital Inc. that covers the high-throughput screening of TCR genes to predict CTCL progression and prognosis. T.S.K. serves on the Scientific Advisory Board (Hematology) of Adaptive Biotechnologies but does not own stock or receive compensation.
View Abstract

Stay Connected to
   Science Translational Medicine

Navigate This Article