`
`CANCER IMMUNOLOGY
`
`Mutational landscape determines
`sensitivity to PD-1 blockade in
`non–small cell lung cancer
`
`Naiyer A. Rizvi,1,2*† Matthew D. Hellmann,1,2* Alexandra Snyder,1,2,3* Pia Kvistborg,4
`Vladimir Makarov,3 Jonathan J. Havel,3 William Lee,5 Jianda Yuan,6 Phillip Wong,6
`Teresa S. Ho,6 Martin L. Miller,7 Natasha Rekhtman,8 Andre L. Moreira,8
`Fawzia Ibrahim,1 Cameron Bruggeman,9 Billel Gasmi,10 Roberta Zappasodi,10
`Yuka Maeda,10 Chris Sander,7 Edward B. Garon,11 Taha Merghoub,1,10
`Jedd D. Wolchok,1,2,10 Ton N. Schumacher,4 Timothy A. Chan2,3,5‡
`
`Immune checkpoint inhibitors, which unleash a patient’s own T cells to kill tumors, are
`revolutionizing cancer treatment. To unravel the genomic determinants of response
`to this therapy, we used whole-exome sequencing of non–small cell lung cancers treated
`with pembrolizumab, an antibody targeting programmed cell death-1 (PD-1). In two
`independent cohorts, higher nonsynonymous mutation burden in tumors was associated
`with improved objective response, durable clinical benefit, and progression-free survival.
`Efficacy also correlated with the molecular smoking signature, higher neoantigen
`burden, and DNA repair pathway mutations; each factor was also associated with mutation
`burden. In one responder, neoantigen-specific CD8+ T cell responses paralleled tumor
`regression, suggesting that anti–PD-1 therapy enhances neoantigen-specific T cell
`reactivity. Our results suggest that the genomic landscape of lung cancers shapes
`response to anti–PD-1 therapy.
`
`T oday, more than a century since the initial
`
`observation that the immune system can re-
`ject human cancers (1), immune checkpoint
`inhibitors are demonstrating that adaptive
`immunity can be harnessed for the treat-
`ment of cancer (2–7). In advanced non–small cell
`lung cancer (NSCLC), therapies with an antibody
`targeting programmed cell death-1 (anti–PD-1) dem-
`onstrated response rates of 17 to 21%, with some
`responses being remarkably durable (3, 8).
`Understanding the molecular determinants of
`response to immunotherapies such as anti–PD-1
`therapy is one of the critical challenges in oncol-
`ogy. Among the best responses have been in
`melanomas and NSCLCs, cancers largely caused
`by chronic exposure to mutagens [ultraviolet light
`
`1Department of Medicine, Memorial Sloan Kettering Cancer
`Center, New York, NY 10065, USA. 2Weill Cornell Medical
`College, New York, NY, 10065, USA. 3Human Oncology and
`Pathogenesis Program, Memorial Sloan Kettering Cancer
`Center, New York, NY 10065, USA. 4Division of Immunology,
`Netherlands Cancer Institute, 1066 CX Amsterdam,
`Netherlands. 5Department of Radiation Oncology, Memorial
`Sloan Kettering Cancer Center, New York, NY 10065, USA.
`6Immune Monitoring Core, Ludwig Center for Cancer
`Immunotherapy, Memorial Sloan Kettering Cancer Center,
`New York, NY 10065, USA. 7Computation Biology Program,
`Memorial Sloan Kettering Cancer Center, New York, NY
`10065, USA. 8Department of Pathology, Memorial Sloan
`Kettering Cancer Center, New York, NY 10065, USA.
`9Department of Mathematics, Columbia University, New
`York, NY, 10027, USA. 10Ludwig Collaborative Laboratory,
`Memorial Sloan Kettering Cancer Center, New York, NY
`10065, USA. 11David Geffen School of Medicine at UCLA,
`2825 Santa Monica Boulevard, Suite 200, Santa Monica, CA
`90404, USA.
`*These authors contributed equally to this work. †Present address:
`Division of Hematology/Oncology, New York-Presbyterian/Columbia
`University, New York, NY, USA. ‡Corresponding author. E-mail:
`chant@mskcc.org
`
`pembrolizumab. In the discovery cohort (n = 16),
`the median number of nonsynonymous muta-
`tions was 302 in patients with durable clinical
`benefit (DCB) (partial or stable response lasting
`>6 months) versus 148 with no durable benefit
`(NDB) (Mann-Whitney P = 0.02) (Fig. 1A). Seventy-
`three percent of patients with high nonsynon-
`ymous burden (defined as above the median
`burden of the cohort, 209) experienced DCB, com-
`pared with 13% of those with low mutation bur-
`den (below median) (Fisher’s exact P = 0.04). Both
`confirmed objective response rate (ORR) and
`progression-free survival (PFS) were higher in
`patients with high nonsynonymous burden [ORR
`63% versus 0%, Fisher’s exact P = 0.03; median
`PFS 14.5 versus 3.7 months, log-rank P = 0.01;
`hazard ratio (HR) 0.19, 95% confidence interval
`(CI) 0.05 to 0.70] (Fig. 1B and table S2).
`The validation cohort included an independent
`set of 18 NSCLC samples from patients treated
`with pembrolizumab. The clinical characteristics
`were similar in both cohorts. The median non-
`synonymous mutation burden was 244 in tu-
`mors from patients with DCB compared to 125
`in those with NDB (Mann-Whitney P = 0.04)
`(Fig. 1C). The rates of DCB and PFS were again sig-
`nificantly greater in patients with a nonsynon-
`ymous mutation burden above 200, the median
`of the validation cohort (DCB 83% versus 22%,
`Fisher’s exact P = 0.04; median PFS not reached
`versus 3.4 months, log-rank P = 0.006; HR 0.15,
`95% CI 0.04 to 0.59) (Fig. 1D and table S2).
`In the discovery cohort, there was high con-
`cordance between nonsynonymous mutation bur-
`den and DCB, with an area under the receiver
`operator characteristic (ROC) curve (AUC) of 87%
`(Fig. 1E). Patients with nonsynonymous muta-
`tion burden ≥178, the cut point that combined
`maximal sensitivity with best specificity, had a
`likelihood ratio for DCB of 3.0; the sensitivity
`and specificity of DCB using this cut point was
`100% (95% CI 59 to 100%) and 67% (29 to 93%),
`respectively. Applying this cut point to the
`validation cohort, the rate of DCB in patients
`with tumors harboring ≥178 mutations was 75%
`compared to 14% in those with <178, corre-
`sponding to a sensitivity of 86% and a specific-
`ity of 75%.
`There were few but important exceptions. Five
`of 18 tumors with ≥178 nonsynonymous muta-
`tions had NDB, and one tumor with a very low
`burden (56 nonsynonymous mutations) responded
`to pembrolizumab. However, this response was
`transient, lasting 8 months. Across both cohorts,
`this was the only patient with a tumor mutation
`burden <178 and confirmed objective response.
`Notably, although higher nonsynonymous mu-
`tation burden correlated with improved ORR,
`DCB, and PFS (Fig. 1, F and G), this correlation
`was less evident when examining total exonic
`mutation burden (table S2).
`We next examined all 34 exomes collectively to
`determine how patterns of mutational changes
`were associated with clinical benefit to pembro-
`lizumab (tables S4 and S5). C-to-A transversions
`were more frequent, and C-to-T transitions were
`less frequent, in patients with DCB compared to
`
`(9) and carcinogens in cigarette smoke (10), re-
`spectively]. However, there is a large variability
`in mutation burden within tumor types, ranging
`from 10s to 1000s of mutations (11–13). This range
`is particularly broad in NSCLCs because tumors
`in never-smokers generally have few somatic mu-
`tations compared with tumors in smokers (14).
`We hypothesized that the mutational landscape
`of NSCLCs may influence response to anti–PD-1
`therapy. To examine this hypothesis, we sequenced
`the exomes of NSCLCs from two independent
`cohorts of patients treated with pembrolizumab,
`a humanized immunoglobulin G (IgG) 4-kappa
`isotype antibody to PD-1 (n = 16 and n = 18, re-
`spectively), and their matched normal DNA (fig.
`S1 and table S1) (15).
`Overall, tumor DNA sequencing generated mean
`target coverage of 164x, and a mean of 94.5% of
`the target sequence was covered to a depth of at
`least 10x; coverage and depth were similar be-
`tween cohorts, as well as between those with or
`without clinical benefit (fig. S2). We identified a
`median of 200 nonsynonymous mutations per
`sample (range 11 to 1192). The median number of
`exonic mutations per sample was 327 (range 45
`to 1732). The quantity and range of mutations were
`similar to published series of NSCLCs (16, 17)
`(fig. S3). The transition/transversion ratio (Ti/Tv)
`was 0.74 (fig. S4), also similar to previously de-
`scribed NSCLCs (16–18). To ensure accuracy of our
`sequencing data, targeted resequencing with an
`orthogonal method (Ampliseq) was performed
`using 376 randomly selected variants, and muta-
`tions were confirmed in 357 of those variants (95%).
`Higher somatic nonsynonymous mutation
`burden was associated with clinical efficacy of
`
`124
`
`3 APRIL 2015 • VOL 348 ISSUE 6230
`
`sciencemag.org SCIENCE
`
`Corrected 11 February 2016; see full text.
`
`Genome Ex. 1029
`Page 1 of 5
`
`
`
`RESEARCH | REPORTS
`
`NDB (Mann-Whitney P = 0.01 for both) (fig. S5).
`A previously validated binary classifier to identi-
`fy the molecular signature of smoking (17) was
`applied to differentiate transversion-high (TH,
`smoking signature) from transversion-low (TL,
`never-smoking signature) tumors. Efficacy was
`greatest in patients with tumors harboring the
`smoking signature. The ORR in TH tumors was
`56% versus 17% in TL tumors (Fisher’s exact P =
`0.03); the rate of DCB was 77% versus 22% (Fisher’s
`exact P = 0.004); the PFS was also significantly
`longer in TH tumors (median not reached versus
`3.5 months, log-rank P = 0.0001) (Fig. 2A). Self-
`reported smoking history did not significantly
`discriminate those most likely to benefit from
`pembrolizumab. The rates of neither DCB nor
`PFS were significantly different in ever-smokers
`versus never-smokers (Fisher’s exact P = 0.66 and
`log-rank P = 0.29, respectively) or heavy smokers
`(median pack-years >25) versus light/never smokers
`(pack-years ≤25) (Fisher’s exact P = 0.08 and log-
`rank P = 0.15, respectively). The molecular smoking
`signature correlated more significantly with non-
`
`synonymous mutation burden than smoking his-
`tory (fig. S6, A and B).
`Although carcinogens in tobacco smoke are
`largely responsible for the mutagenesis in lung
`cancers (19), the wide range of mutation burden
`within both smokers and never-smokers impli-
`cates additional pathways contributing to the
`accumulation of somatic mutations. We found
`deleterious mutations in a number of genes that
`are important in DNA repair and replication. For
`example, in three responders with the highest
`mutation burden, we identified deleterious mu-
`tations in POLD1, POLE, and MSH2 (Fig. 3). Of
`particular interest, a POLD1 E374K mutation was
`identified in a never-smoker with DCB whose tu-
`mor harbored the greatest nonsynonymous muta-
`tion burden (n = 507) of all never-smokers in our
`series. POLD1 Glu374 lies in the exonuclease proof-
`reading domain of Pol d (20), and mutation of
`this residue may contribute to low-fidelity repli-
`cation of the lagging DNA strand. Consistent with
`this hypothesis, this tumor exome had a relatively
`low proportion of C-to-A transversions (20%) and
`
`predominance of C-to-T transitions (51%), similar
`to other POLD1 mutant, hypermutated tumors
`(21) and distinct from smoking-related lung can-
`cers. Another responder, with the greatest muta-
`tion burden in our series, had a C284Y mutation
`in POLD1, which is also located in the exonu-
`clease proofreading domain. We observed non-
`sense mutations in PRKDC, the catalytic subunit
`of DNA-dependent protein kinase (DNA-PK),
`and RAD17. Both genes are required for proper
`DNA repair and maintenance of genomic integ-
`rity (22, 23).
`Genes harboring deleterious mutations com-
`mon to four or more DCB patients and not present
`in NDB patients included POLR2A, KEAP1, PAPPA2,
`PXDNL, RYR1, SCN8A, and SLIT3. Mutations in
`KRAS were found in 7 of 14 tumors from patients
`with DCB compared to 1 of 17 in the NDB group,
`a finding that may be explained by the asso-
`ciation between smoking and the presence of
`KRAS mutations in NSCLC (24). There were no
`mutations or copy-number alterations in antigen-
`presentation pathway–associated genes or CD274
`
`Discovery Cohort
`
`Validation Cohort
`
`All Tumors
`
`DCB
`
`NDB
`
`1200
`1 2 0 0
`
`800
`8 0 0
`
`400
`4 0 0
`
`200
`2 0 0
`
` 0
`0
`
`# Nonsynonymous mutations/tumor
`
`50
`5 0
`1 - % Specificity
`
`100
`1 0 0
`
`100
`1 0 0
`
`50
`5 0
`
`% Sensitivity
`
` 0
`0
`0
`
`DCB
`
`NDB
`
`1200
`1 2 0 0
`
`800
`8 0 0
`
`400
`4 0 0
`
`200
`2 0 0
`
` 0
`0
`
`# Nonsynonymous mutations/tumor
`
`DCB
`
`NDB
`
`800
`8 0 0
`
`600
`6 0 0
`
`400
`4 0 0
`
`200
`2 0 0
`
` 0
`0
`
`# Nonsynonymous mutations/tumor
`
`Discovery Cohort
`
`Validation Cohort
`
`All Tumors
`
`High nonsynonymous burden
`Low nonsynonymous burden
`
`100
`1 0 0
`
`50
`5 0
`
`Percent progression-free
`
`High nonsynonymous burden
`Low nonsynonymous burden
`
`High nonsynonymous burden
`Low nonsynonymous burden
`
`100
`1 0 0
`
`50
`5 0
`
`Percent progression-free
`
`100
`1 0 0
`
`50
`5 0
`
`Percent progression-free
`
` 0
`0
`0
`
` 4
`4
`
` 8
`8
`
` 12
` 16
`1 2
`1 6
`Months
`
` 20
`2 0
`
` 24
`2 4
`
` 0
`0
`
`0
`
` 4
`4
`
` 8
`8
`
` 12
` 16
`1 2
`1 6
`Months
`
` 20
`2 0
`
` 24
`2 4
`
` 0
`0
`0
`
` 4
`4
`
` 8
`8
`
` 20
`2 0
`
` 24
`2 4
`
`Fig. 1. Nonsynonymous mutation burden associated with clinical bene-
`fit of anti–PD-1 therapy. (A) Nonsynonymous mutation burden in tumors
`from patients with DCB (n = 7) or with NDB (n = 9) (median 302 versus
`148, Mann-Whitney P = 0.02). (B) PFS in tumors with higher nonsynony-
`mous mutation burden (n = 8) compared to tumors with lower nonsynony-
`mous mutation burden (n = 8) in patients in the discovery cohort (HR 0.19,
`95% CI 0.05 to 0.70, log-rank P = 0.01). (C) Nonsynonymous mutation
`burden in tumors with DCB (n = 7) compared to those with NDB (n = 8) in
`patients in the validation cohort (median 244 versus 125, Mann-Whitney
`P = 0.04). (D) PFS in tumors with higher nonsynonymous mutation burden
`(n = 9) compared to those with lower nonsynonymous mutation burden
`(n = 9) in patients in the validation cohort (HR 0.15, 95% CI 0.04 to 0.59,
`
`SCIENCE sciencemag.org
`
`3 APRIL 2015 • VOL 348 ISSUE 6230
`
`125
`
` 12
` 16
`1 2
`1 6
`Months
`log-rank P = 0.006). (E) ROC curve for the correlation of nonsynonymous
`mutation burden with DCB in discovery cohort. AUC is 0.86 (95% CI 0.66
`to 1.05, null hypothesis test P = 0.02). Cut-off of ≥178 nonsynonymous mu-
`tations is designated by triangle. (F) Nonsynonymous mutation burden in
`patients with DCB (n = 14) compared to those with NDB (n = 17) for the
`entire set of sequenced tumors (median 299 versus 127, Mann-Whitney P =
`0.0008). (G) PFS in those with higher nonsynonymous mutation burden
`(n = 17) compared to those with lower nonsynonymous mutation burden
`(n = 17) in the entire set of sequenced tumors (HR 0.19, 95% CI 0.08-0.47,
`log-rank P = 0.0004). In (A), (C), and (F), median and interquartile ranges of
`total nonsynonymous mutations are shown, with individual values for each
`tumor shown with dots.
`
`Corrected 11 February 2016; see full text.
`
`Genome Ex. 1029
`Page 2 of 5
`
`
`
`How does increased mutation burden affect tu-
`mor immunogenicity? The observation that non-
`synonymous mutation burden is associated with
`pembrolizumab efficacy is consistent with the
`hypothesis that recognition of neoantigens, formed
`as a consequence of somatic mutations, is impor-
`tant for the activity of anti–PD-1 therapy. We ex-
`amined the landscape of neoantigens using our
`previously described methods (25) (fig. S7). Briefly,
`this approach identifies mutant nonamers with
`≤500 nM binding affinity for patient-specific class
`I human lymphocyte antigen (HLA) alleles (26, 27),
`which are considered candidate neoantigens (table
`S6). We identified a median of 112 candidate neo-
`antigens per tumor (range 8 to 610), and the quan-
`tity of neoantigens per tumor correlated with
`mutation burden (Spearman r 0.91, P < 0.0001),
`similar to the correlation recently reported across
`cancers (28). Tumors from patients with DCB had
`significantly higher candidate neoantigen bur-
`den compared to those with NDB (Fig. 4A), and
`high candidate neoantigen burden was associated
`with improved PFS (median 14.5 versus 3.5 months,
`log-rank P = 0.002) (Fig. 4B). The presence of sp-
`
`ecific HLA alleles did not correlate with efficacy
`(fig. S8). The absolute burden of candidate neo-
`antigens, but not the frequency per nonsynony-
`mous mutation, correlated with response (fig. S9).
`We next sought to assess whether anti–PD-1
`therapy can alter neoantigen-specific T cell re-
`activity. To directly test this, identified candidate
`neoantigens were examined in a patient (Study
`ID no. 9 in Fig. 3 and table S3) with exceptional
`response to pembrolizumab and available pe-
`ripheral blood lymphocytes (PBLs). Predicted
`HLA-A–restricted peptides were synthesized to
`screen for ex vivo autologous T cell reactivity in
`serially collected PBLs (days 0, 21, 44, 63, 256, and
`297, where day 0 is the first date of treatment)
`using a validated high-throughput major histo-
`compatibility complex (MHC) multimer screening
`strategy (29, 30). This analysis revealed a CD8+
`T cell response against a neoantigen resulting
`from a HERC1 P3278S mutation (ASNAS SAAK)
`(Fig. 4C). Notably, this T cell response could only
`be detected upon the start of therapy (level of
`detection 0.005%). Three weeks after therapy
`initiation, the magnitude of response was 0.040%
`
`RESEARCH | REPORTS
`
`[encoding programmed cell death ligand-1 (PD-L1)]
`that were associated with response or resistance.
`
`Transversion high
`Transversion low
`
`100
`
`1 0 0
`
`50
`
`5 0
`
`Percent progression-free
`
` 0
`
`0
`
`0
`
`4
` 4
`
`8
` 8
`
`20
` 20
`
`24
` 24
`
`12
`16
` 12
` 16
`Months
`Fig. 2. Molecular smoking signature is significantly
`associated with improved PFS in NSCLC patients
`treated with pembrolizumab. PFS in tumors char-
`acterized as TH by molecular smoking signature
`classifier (n = 16) compared to TL tumors (n = 18)
`(HR 0.15, 95% 0.06 to 0.39, log-rank P = 0.0001).
`
`Fig. 3. Mutation burden, clinical response, and factors contributing to
`mutation burden.Total exonic mutation burden for each sequenced tumor with
`nonsynonymous (dark shading), synonymous (medium shading), and indels/
`frameshift mutations (light shading) displayed in the histogram. Columns are
`shaded to indicate clinical benefit status: DCB, green; NDB, red; not reached
`6 months follow-up (NR), blue. The cohort identification (D, discovery; V, valida-
`
`tion), best objective response (PR, partial response; SD, stable disease; PD,
`progression of disease), and PFS (censored at the time of data lock) are reported
`in the table.Those with ongoing progression-free survival are labeled with ++.The
`presence of the molecular smoking signature is displayed in the table with TH
`cases (purple) and TL cases (orange). The presence of deleterious mutations in
`specific DNA repair/replication genes is indicated by the arrows.
`
`126
`
`3 APRIL 2015 • VOL 348 ISSUE 6230
`
`sciencemag.org SCIENCE
`
`Corrected 11 February 2016; see full text.
`
`Genome Ex. 1029
`Page 3 of 5
`
`
`
`RESEARCH | REPORTS
`
`of CD8+ T cells, and this response was main-
`tained at Day 44. This rapid induction of T cell
`reactivity correlated with tumor regression, and
`this T cell response returned to levels just above
`background in the subsequent months as tumor
`regression plateaued (Fig. 4D). HERC1 P3278S-
`multimer–reactive T cells from PBLs collected on
`day 44 were characterized by a CD45RA-CCR7-
`HLA-DR+LAG-3 phenotype, consistent with an
`activated effector population (fig. S10). These data
`reveal autologous T cell responses against cancer
`neoantigens in the context of a clinical response
`to anti–PD-1 therapy.
`To validate the specificity of the neoantigen-
`reactive T cells, PBLs from days 63 and 297 were
`expanded in vitro in the presence of mutant pep-
`tide and subsequently restimulated with either
`mutant or wild-type peptide (ASNASSAAK versus
`
`ASNAPSAAK), and intracellular cytokines were
`analyzed. At both time points, a substantial pop-
`ulation of polyfunctional CD8+ T cells [charac-
`terized by production of the cytokines interferon
`(IFN) g and tumor necrosis factor (TNF) a, the
`marker of cytotoxic activity CD107a, and the chemo-
`kine CCL4] was detected in response to mutant
`but not wild-type peptide (Fig. 4E and fig. S11).
`In the current study, we show that in NSCLCs
`treated with pembrolizumab, elevated nonsynon-
`ymous mutation burden strongly associates with
`clinical efficacy. Additionally, clinical efficacy cor-
`relates with a molecular signature characteristic
`of tobacco carcinogen–related mutagenesis, cer-
`tain DNA repair mutations, and the burden of
`neoantigens. The molecular smoking signature
`correlated with efficacy, whereas self-reported
`smoking status did not, highlighting the power
`
`of this classifier to identify molecularly related
`tumors within a heterogeneous group.
`Previous studies have reported that pretreat-
`ment PD-L1 expression enriches for response to
`anti–PD-1 therapies (3, 8, 31), but many tumors
`deemed PD-L1 positive do not respond, and some
`responses occur in PD-L1–negative tumors (8, 31).
`Semiquantitative PD-L1 staining results were avail-
`able for 30 of 34 patients, where strong staining
`represented ≥50% PD-L1 expression, weak rep-
`resented 1 to 49%, and negative represented
`<1% [clone 22C3, Merck (8)]. As this trial largely
`enrolled patients with PD-L1 tumor expression,
`most samples had some degree of PD-L1 ex-
`pression (24 of 30, 80%) (table S3), limiting the
`capacity to determine relationships between
`mutation burden and PD-L1 expression. Among
`those with high nonsynonymous mutation burden
`
`Qdot 625 pMHC multimer
`
`No Stimulation
`WT-ASNAPSAAK
`MUT-ASNASSAAK
`
`Day
`63
`
`Day
`297
`
`BV421 pMHC multimer
`
`IFN
`
`High neoantigen burden
`Low neoantigen burden
`
` 4
`
` 8
`
` 12 16 20
`Months
`
` 24
`
`100
`
`50
`
`Percent progression-free
`
` 0
` 0
`
`DCB
`
`NDB
`
`Baseline
`
`Day 56
`
`Day 315
`
`600
`6 0 0
`
`400
`4 0 0
`
`200
`2 0 0
`
` 0
`0
`
`# Candidate neoantigens/tumor
`
`Liver metastasis
`
`0
`
`CD8
`
`TNF
`
`CD107a
`
`CCL4
`
`-78%
`
`-88%
`
`-89%
`
`-88%
`
`-91%
`
`Day 44, 0.044%
`
`Day 21,
`0.04%
`
`Day 63, 0.022%
`
`Pretreatment Day 1, 0.001%
`
`Day 256,
`0.003%
`
`Day 297,
`0.005%
`
`-50
`
`-100
`
`0.05
`
`0.04
`
`0.03
`
`0.02
`
`0.01
`
`0.00
`
`response (irRC)
`
`Clinical
`
`HERC1 P3278S multimers
`
`% PBLs reactive to HLA-A*11:01
`
`Fig. 4. Candidate neoantigens, neoantigen-specific T cell
`response, and response to pembrolizumab. (A) Neoantigen
`burden in patients with DCB (n = 14) compared to NDB (n = 17)
`across the overall set of sequenced tumors (median 203 versus
`83, Mann-Whitney P = 0.001). (B) PFS in tumors with higher can-
`didate neoantigen burden (n = 17) compared to tumors with lower
`candidate neoantigen burden (n = 17) (HR 0.23, 95% CI 0.09 to
`0.58, log-rank P = 0.002). (C) (Top) Representative computed
`tomography (CT) images of a liver metastasis before and after
`initiation of treatment. (Middle) Change in radiographic response.
`(Bottom) Magnitude of the HERC1 P3278S reactive CD8+ T cell
`response measured in peripheral blood. (D) The proportion of
`CD8+ T cell population in serially collected autologous PBLs rec-
`ognizing the HERC1 P3278S neoantigen (ASNASSAAK) before and
`0 60 120 180 240 300
`Days
`during pembrolizumab treatment. Each neoantigen is encoded
`by a unique combination of two fluorescently labeled peptide-
`MHC complexes (represented individually on each axis); neoantigen-specific T cells are represented by the events in the double positive position indicated
`with black dots. Percentages indicate the number of CD8+ MHC multimer+ cells out of total CD8 cells. (E) Autologous T cell response to wild-type HERC1
`peptide (black), mutant HERC1 P3278S neoantigen (red), or no stimulation (blue), as detected by intracellular cytokine staining. T cell costains for IFNg and
`CD8, TNFa, CD107a, and CCL4, respectively, are displayed for the Day 63 and Day 297 time points.
`
`SCIENCE sciencemag.org
`
`3 APRIL 2015 • VOL 348 ISSUE 6230
`
`127
`
`Corrected 11 February 2016; see full text.
`
`Genome Ex. 1029
`Page 4 of 5
`
`
`
`RESEARCH | REPORTS
`
`(>200, above median of overall cohort) and some
`degree of PD-L1 expression (weak/strong), the
`rate of DCB was 91% (10 of 11, 95% CI 59 to
`99%). In contrast, in those with low mutation
`burden and some degree of PD-L1 expression,
`the rate of DCB was only 10% (1 of 10, 95% CI
`0 to 44%). When exclusively examining patients
`with weak PD-L1 expression, high nonsynonymous
`mutation burden was associated with DCB in
`75% (3 of 4, 95% CI 19 to 99%), and low mutation
`burden was associated with DCB in 11% (1 of 9,
`0 to 48%). Large-scale studies are needed to deter-
`mine the relationship between PD-L1 intensity
`and mutation burden. Additionally, recent data
`have demonstrated that the localization of PD-L1
`expression within the tumor microenvironment
`[on infiltrating immune cells (32), at the invasive
`margin, tumor core, and so forth (33)] may affect
`the use of PD-L1 as a biomarker.
`T cell recognition of cancers relies upon pre-
`sentation of tumor-specific antigens on MHC
`molecules (34). A few preclinical (35–41) and clin-
`ical reports have demonstrated that neoantigen-
`specific effector T cell response can recognize
`(25, 42–45) and shrink established tumors (46).
`Our finding that nonsynonymous mutation bur-
`den more closely associates with pembrolizumab
`clinical benefit than total exonic mutation burden
`suggests the importance of neoantigens in dic-
`tating response.
`The observation that anti–PD-1–induced
`neoantigen-specific T cell reactivity can be ob-
`served within the peripheral blood compartment
`may open the door to development of blood-
`based assays to monitor response during anti–
`PD-1 therapy. We believe that our findings have
`an important impact on our understanding of re-
`sponse to anti–PD-1 therapy and on the applica-
`tion of these agents in the clinic.
`
`REFERENCES AND NOTES
`1. W. B. Coley, Clin. Orthop. Relat. Res. 1991(262), 3–11 (1991).
`2. F. S. Hodi et al., N. Engl. J. Med. 363, 711–723 (2010).
`3. S. L. Topalian et al., N. Engl. J. Med. 366, 2443–2454 (2012).
`J. D. Wolchok et al., N. Engl. J. Med. 369, 122–133 (2013).
`4.
`5. C. Robert et al., Lancet 384, 1109–1117 (2014).
`6. T. Powles et al., Nature 515, 558–562 (2014).
`7. S. M. Ansell et al., N. Engl. J. Med. 372, 311–319 (2015).
`8. E. B. Garon et al., Ann. Oncol. 25, LBA43 (2014).
`9. G. P. Pfeifer, Y. H. You, A. Besaratinia, Mutat. Res. 571, 19–31
`(2005).
`10. G. P. Pfeifer et al., Oncogene 21, 7435–7451 (2002).
`11. M. S. Lawrence et al., Nature 499, 214–218 (2013).
`12. L. B. Alexandrov et al., Nature 500, 415–421 (2013).
`13. B. Vogelstein et al., Science 339, 1546–1558 (2013).
`14. R. Govindan et al., Cell 150, 1121–1134 (2012).
`15. See supplementary text available on Science Online.
`16. P. S. Hammerman et al., Nature 489, 519–525 (2012).
`17. Cancer Genome Atlas Research Network, Nature 511, 543–550
`(2014).
`18. O. D. Abaan et al., Cancer Res. 73, 4372–4382 (2013).
`19. D. Hoffmann, I. Hoffmann, K. El-Bayoumy, Chem. Res. Toxicol.
`14, 767–790 (2001).
`20. R. Hindges, U. Hübscher, Biol. Chem. 378, 345–362 (1997).
`21. C. Palles et al., Nat. Genet. 45, 136–144 (2013).
`22. J. F. Goodwin, K. E. Knudsen, Cancer Discov. 4, 1126–1139 (2014).
`23. X. Wang et al., Genes Dev. 17, 965–970 (2003).
`24. S. Dogan et al., Clin. Cancer Res. 18, 6169–6177 (2012).
`25. A. Snyder et al., N. Engl. J. Med. 371, 2189–2199 (2014).
`26. M. Nielsen et al., Protein Sci. 12, 1007–1017 (2003).
`27. C. Lundegaard et al., Nucleic Acids Res. 36 (Web Server),
`W509–W512 (2008).
`28. M. S. Rooney, S. A. Shukla, C. J. Wu, G. Getz, N. Hacohen,
`Cell 160, 48–61 (2015).
`
`29. B. Rodenko et al., Nat. Protoc. 1, 1120–1132 (2006).
`30. R. S. Andersen et al., Nat. Protoc. 7, 891–902 (2012).
`31. J. M. Taube et al., Clin. Cancer Res. 20, 5064–5074 (2014).
`32. R. S. Herbst et al., Nature 515, 563–567 (2014).
`33. P. C. Tumeh et al., Nature 515, 568–571 (2014).
`34. R. D. Schreiber, L. J. Old, M. J. Smyth, Science 331, 1565–1570
`(2011).
`35. T. Matsutake, P. K. Srivastava, Proc. Natl. Acad. Sci. U.S.A. 98,
`3992–3997 (2001).
`36. H. Matsushita et al., Nature 482, 400–404 (2012).
`37. J. C. Castle et al., Cancer Res. 72, 1081–1091 (2012).
`38. T. Schumacher et al., Nature 512, 324–327 (2014).
`39. M. M. Gubin et al., Nature 515, 577–581 (2014).
`40. M. Yadav et al., Nature 515, 572–576 (2014).
`41. F. Duan et al., J. Exp. Med. 211, 2231–2248 (2014).
`42. N. van Rooij et al., J. Clin. Oncol. 31, e439–e442 (2013).
`43. P. F. Robbins et al., Nat. Med. 19, 747–752 (2013).
`44. M. Rajasagi et al., Blood 124, 453–462 (2014).
`45. C. Linnemann et al., Nat. Med. 21, 81–85 (2015).
`46. E. Tran et al., Science 344, 641–645 (2014).
`
`ACKNOWLEDGMENTS
`We thank the members of the Thoracic Oncology Service and
`the Chan and Wolchok laboratories at Memorial Sloan Kettering
`Cancer Center (MSKCC) for helpful discussions. We thank the
`Immune Monitoring Core at MSKCC, including L. Caro, R. Ramsawak,
`and Z. Mu, for exceptional support with processing and banking
`peripheral blood lymphocytes. We thank P. Worrell and E. Brzostowski
`for help in identifying tumor specimens for analysis. We thank
`A. Viale for superb technical assistance. We thank D. Philips,
`M. van Buuren, and M. Toebes for help performing the combinatorial
`
`coding screens. The data presented in this paper are tabulated in
`the main paper and in the supplementary materials. Data are publicly
`available at the Cancer Genome Atlas (TCGA) cBio portal and
`database (www.cbioportal.org; study ID: Rizvi lung cancer). All
`genotype and phenotype data are deposited at dpGAP under
`accession no. phs000980.v1.p1. T.A.C. is the inventor on a patent
`(provisional application number 62/083,088). The application is
`directed toward methods for identifying patients who will benefit from
`treatment with immunotherapy. This work was supported by the
`Geoffrey Beene Cancer Research Center (M.D.H., N.A.R., T.A.C., J.D.W.,
`and A.S.), the Society for Memorial Sloan Kettering Cancer Center
`(M.D.H.), Lung Cancer Research Foundation (W.L.), Frederick Adler
`Chair Fund (T.A.C.), The One Ball Matt Memorial Golf Tournament
`(E.B.G.), Queen Wilhelmina Cancer Research Award (T.N.S.), The
`STARR Foundation (T.A.C. and J.D.W.), the Ludwig Trust (J.D.W.),
`and a Stand Up To Cancer-Cancer Research Institute Cancer
`Immunology Translational Cancer Research Grant (J.D.W., T.N.S.,
`and T.A.C.). Stand Up To Cancer is a program of the Entertainment
`Industry Foundation administered by the American Association for
`Cancer Research.
`
`SUPPLEMENTARY MATERIALS
`www.sciencemag.org/content/348/6230/124/suppl/DC1
`Materials and Methods
`Figs. S1 to S12
`Tables S1 to S6
`References (47–68)
`
`21 October 2014; accepted 27 February 2015
`Published online 12 March 2015;
`10.1126/science.aaa1348
`
`GENE EXPRESSION
`
`MicroRNA control of protein
`expression noise
`
`Jörn M. Schmiedel,1,2,3 Sandy L. Klemm,4 Yannan Zheng,3 Apratim Sahay,3
`Nils Blüthgen,1,2*† Debora S. Marks,5*† Alexander van Oudenaarden3,6,7*†
`
`MicroRNAs (miRNAs) repress the expression of many genes in metazoans by accelerating
`messenger RNA degradation and inhibiting translation, thereby reducing the level of protein.
`However, miRNAs only slightly reduce the mean expression of most targeted proteins, leading
`to speculation about their role in the variability, or noise, of protein expression. We used
`mathematical modeling and single-cell reporter assays to show that miRNAs, in conjunction
`with increased transcription, decrease protein expression noise for lowly expressed genes
`but increase noise for highly expressed genes. Genes that are regulated by multiple miRNAs
`show more-pronounced noise reduction. We estimate that hundreds of (lowly expressed)
`genes in mouse embryonic stem cells have reduced noise due to substantial miRNA regulation.
`Our findings suggest that miRNAs confer precision to protein expression and thus offer
`plausible explanations for the commonly observed combinatorial targeting of endogenous genes
`by multiple miRNAs, as well as the preferential targeting of lowly expressed genes.
`
`M icroRNAs (miRNAs) regulate numerous
`
`genes in metazoan organisms (1–5) by
`accelerating mRNA degradation and
`inhibiting translation (6, 7). Although the
`physiological function of some miRNAs
`is known in detail (1, 2, 8, 9), it is unclear why
`miRNA regulation is so ubiquitous and conserved,
`because individual miRNAs only weakly repress
`the vast majority of their target genes (10, 11), and
`knockouts rarely show phenotypes (12). One
`proposed reason for this widespread regulation
`is the ability of miRNAs to provide precision to
`gene expression (13). Previous work has hy-
`pothesized that miRNAs could reduce protein
`expression variability (noise) when their repres-
`
`sive posttranscriptional effects are antagonized
`by accelerated transcriptional dynamics (14, 15).
`However, because miRNA levels are themselves
`variable, one should expect the propagation of
`their fluctuations to introduce additional noise
`(Fig. 1A).
`To test the effects of endogenous miRNAs, we
`quantified protein levels and fluctuations in
`mouse embryonic stem cells (mESCs) using a
`dual fluorescent reporter system (16), in which
`two different reporters (ZsGreen and mCherry)
`a