Abstract
The majority of lung cancer patients are refractory to PD-L1/PD-1 blockade monotherapy. This therapy may even accelerate progression and death in a group of patients called hyperprogressors. Here we demonstrate that the efficacy of PD-L1/PD-1 blockade therapy relies on baseline circulating highly-differentiated CD28− CD27− CD4 T cells (THD cells), which segregate patients in two non-overlapping groups. THD cells in cancer patients mostly comprised of central memory subsets that potently co-upregulated PD-1 and LAG3 upon antigen recognition. Low baseline THD numbers unequivocally identified intrinsic non-responders and hyperprogressors, whom aberrantly responded to therapy with a potent systemic proliferative THD cell burst. Responder patients showed significant reductions in systemic CD4 THD cells throughout therapy linked to expansion of the CD28+ CD27+ CD4 T cell compartment. Quantification of THD cells from peripheral blood samples prior to therapy allows identification of non-responders, hyperprogressors and responders, a critical issue in clinical oncology. These results place CD4 T cell responses at the center of anti-tumor immunity.
Introduction
PD-L1/PD-1 blockade is demonstrating remarkable clinical outcomes since its first clinical application in human therapy (Brahmer et al., 2012; Topalian et al., 2012). These therapies interfere with immunosuppressive PD-L1/PD-1 interactions by systemic administration of blocking antibodies. PD-L1 is frequently overexpressed by tumor cells and correlates with progression and resistance to pro-apoptotic stimuli (Azuma et al., 2008; Gato-Canas et al., 2017; Juneja et al., 2017). PD-1 is expressed in antigen-experienced T cells and interferes with T cell activation when engaged with PD-L1 (Chemnitz et al., 2004; Karwacz et al., 2011).
A significant number of lung cancer patients are intrinsically refractory to PD-L1/PD-1 immune checkpoint blockade therapies. Indeed, there is accumulating evidence that PD-L1/PD-1 blockade might even have deleterious effects by accelerating disease and death in a group of patients called hyperprogressors (Champiat et al., 2017; Saada-Bouzid et al., 2017). Currently, there is no way of identifying them before the start of therapy. Hence, stratification of patients into hyperprogressors, non-responders and potential responders is of critical importance (Topalian et al., 2016) especially when the relevance of tumor PD-L1 expression as a predictive biomarker is still under debate, at least for some cancer types (Grigg and Rizvi, 2016). Other biomarkers such as the immunoscore (Galon et al., 2006), interferon gene signatures, mutational load and microsatellite instability require relatively large biopsies (Jamieson and Maker, 2017; Vranic, 2017). These techniques are often difficult to implement in a clinical context, and the retrieval of representative biopsies can sometimes be challenging.
In addition, the impact of T cell terminal differentiation in aged patients is often overlooked in cancer immunotherapies. Highly differentiated human T cells (THD) accumulate with age by a progressive loss of CD27 and CD28 expression (Lanna et al., 2014). T cells are classified as poorly differentiated (CD28+ CD27+ TPD cells), intermediately differentiated cells which lose CD27 expression (CD28+ CD27-TINT cells) and THD cells (CD28-CD27-). In most cases THD cells correspond to terminally differentiated T cells with effector activities, also known as EMRA or senescent cells (Lanna et al., 2017; Lanna et al., 2014) (Fig. 1A). Although dysfunctional to some degree due to reduced expression of CD28 and CD27 co-receptors, these cells do constitute a large pool of antigen-specific T cells with potent effector activities when mobilized (Lanna et al., 2017; Lanna et al., 2014). Indeed, PD-1 blockade can recover some effector activities of CD8 THD cells in vitro (Henson et al., 2015). However, the clinical impact of PD-L1/PD-1 immune checkpoint inhibitor therapy in THD cells and its relationship with objective clinical responses have not been addressed yet. Here we focused on understanding the impact of THD cells over clinical responses in NSCLC patients undergoing PD-L1/PD-1 immune checkpoint blockade therapy.
Results
Baseline CD4 THD cells separate NSCLC patients in two non-overlapping groups
Systemic circulating THD cells were quantified in a cohort of 45 NSCLC patients and healthy age-matched donors (64.8 ± 8.3 vs 68.60 ± 8 years old, mean±standard deviation, SD). Cancer patients showed highly significantly increased numbers of CD4 THD cells (P<0.0001) but not of CD8 THD cells. Importantly, patients were distributed in two non-overlapping groups according to CD4 THD baseline numbers (Fig. 1B); Patients with above-average (G1 patients, 72.2% THD [67.2–76.8, 95% confidence interval (C.I.), N=21]) and below-average numbers (G2 patients, 28.2% THD [24–32.5, 95% C.I., N=24]). Accordingly, NSCLC patients showed very significantly lower numbers of CD28+ CD27+ CD4 T cells (P<0.001) (Fig. 1B). As CD4 THD cells segregated patients in two separate groups, we decided to characterize them at baseline and throughout treatment.
Surprisingly, CD4 TPD, TINT and THD cells consisted of a mixture of differentiation phenotypes, including naïve/stem memory (CD62L+ CD45RA+), central memory (CM, CD62L+ CD45RA−), effector memory (EM, CD62L− CD45RA−) and effector cells (EF, CD62Llow/neg CD45RA+ or EMRA cells), independently on whether they came from G1 or G2 patients (Fig. 1C, 1D). CD4 THD were enriched in cells with central/effector memory phenotypes but not in EMRA cells, suggesting that these CD28− CD27− T cells were not truly terminally differentiated or senescent, although probably dysfunctional. To test this hypothesis, PD-1 expression was evaluated in systemic CD4 THD cells from NSCLC G1 and G2 patients. No significant PD-1 surface expression was observed in these cells (Fig. 1E). This was surprising as constitutive expression of markers such as PD-1 or LAG3 are trademarks of dysfunctional T cells in cancer. However, when CD4 T cells were stimulated by co-incubation with human lung A549 adenocarcinoma cells expressing a membrane-bound anti-CD3 antibody, CD4 T cells from NSCLC patients showed a strong co-up-regulation of PD-1 and LAG3 (Fig. 1E). A different expression pattern was observed without PD-1/LAG-3 co-expression when CD4 T cells from healthy donors were evaluated, in which PD-1 or LAG-3 represent activation markers. Moreover, PD-1 up-regulation was very significantly (P<0.001) pronounced in the more differentiated CD4 subsets, and especially in T cells from NSCLC patients (Fig. 1F).
Baseline CD4 THD numbers discriminate responses to PD-L1/PD-1 immune checkpoint blockade therapies
34 of the NSCLC patients continued with nivolumab, pembrolizumab or atezolizumab treatments following their current indications, and at the end of the study responders accounted to 20% (7 out of 34), consistent with the published efficacies for these agents (Herbst et al., 2016; Horn et al., 2017; Rittmeyer et al., 2017). To evaluate the impact of circulating CD4 THD over immunotherapies, we monitored CD4 THD cell numbers throughout therapy from routine small fresh blood samples. Strikingly, CD4 THD cell values before the start of therapy unequivocally discriminated patients according to responses (P=0.0008) (Fig. 2A). G2 patients (THD values below 40%) were all progressors [19 patients with 26.9% ± 7.8 baseline THD cells, (23–30.8, 95% C.I.)], while responders accounted to 47% of G1 patients with THD values above 40% [7 out of 15 patients with 71.5% ± 9 baseline THD cells, (63–80, 95% C.I.)]. Reciprocally, patients with CD28+ CD27+ CD4 T cell baseline values above 40% were all progressors (Fig. 2B) (P=0.005). Therefore, we defined patients with a “positive” baseline profile as those belonging to G1, while G2 represented patients with a “negative” baseline profile.
Immune checkpoint blockade therapy induces unique dynamic changes in circulating CD4 THD cells that correlate with clinical responses
Immune checkpoint inhibitors strongly affected CD4 T cell populations within the first cycle of treatment, and two main distinct dynamic profiles were identified. Pattern 1 or “THD burst” consisted in a highly significant increase in systemic CD4 THD cells [12.4% increase, (6.2, 18.5) 95% CI, N=27, P<0.0001, one-tailed paired t test)] and was associated to tumor progression without exception in our cohort of patients (Fig. 3A, 3C, 3D). Pattern 2 or THD decrease was characterized by very significant reduction in systemic THD cells [−14.4%, (−8, −21), 95% CI, N=7, P<0.0001], concomitant to an expansion of CD28+ CD27+ CD4 T cells and primarily associated to tumor regression (Fig. 3B, 3C, 3D). There was a very highly significant correlation (P=0.0001) between THD changes and therapeutic outcome (Fig. 3D).
To find out if THD bursts were the result of active proliferation of CD4 THD cells from baseline to post-first cycle of therapy, Ki67 expression was analyzed by intracellular flow cytometry in THD cells from non-responder and responder patients (Fig. 4A, 4B). THD cells in non-responders showed a significant increase in Ki67 expression. In contrast, responders showed a markedly decrease in THD Ki67 expression from baseline to post-first cycle of therapy, together with increased Ki67 expression in the CD28+ CD27+ CD4 T cell compartment (Fig. 4C). These results indicated that THD bursts were likely the result of systemic THD proliferation.
To test whether THD bursts occurred in specific subsets, the changes from baseline to post-first cycle of therapy were studied only in patients exhibiting THD bursts, and compared to those from responder patients. Although the changes were not significant either in responders (with THD decrease) or non-responders exhibiting THD bursts with the current number of patients, there were clear trends. CD4 THD bursts were enriched in CM subsets in detriment of more differentiated effector subsets (EM and EF cells) (Fig. 4D). In contrast, CD4 THD cells that remained in responders post-first cycle of therapy were enriched in EM and EF cells.
CD4 THD bursts define primary resistance to PD-L1/PD-1 blockade and hyperprogression
All G2 patients showed tumor progression (Fig. 5A). Within this group, hyperprogressors were identified following the definition by Sâada-Bouzid et al (Saada-Bouzid et al., 2017) but using as a threshold a tumor growth kinetics ratio equal or superior to 5 (Fig. 5B). We confirmed that a negative THD baseline profile significantly correlated with radiologically-confirmed hyperprogressors (P=0.01) (Fig. 5C) whom showed highly significant THD bursts (P=0.0001) following the first cycle of therapy (Fig. 5D). Six patients were identified as probable hyperprogressors by clinical parameters, whom either died before radiological confirmation or the disease was not evaluable by radiological criteria. Their immunological profiles were consistent with radiologically-confirmed hyperprogressors (Fig. 5C and 5D). All of them experienced early progression of disease compared to the rest (median progression-free survival (mPFS)=6 weeks [5.7–6.3, 95% C.I.] versus 8.9 weeks [4.6–13.1, 95% C.I.], p=0.002).
The agreement between the radiological criterium and the immunological profiling was significant in the identification of hyperprogression by a kappa test (κ=0.742). Hence, a G2 profile associated to significant “THD bursts” objectively characterized hyperprogressive disease in NSCLC patients without being influenced by previous tumor burden or dynamics.
Baseline THD numbers constitute a strong and reliable predictive biomarker for primary resistance and hyperprogression with practical clinical application
To improve specificity even at the cost of a lower sensitivity for identification of intrinsic non-responders, the negative baseline T cell profile was refined as the combination of CD4 THD <40% and CD28+ CD27+ CD4 T cells > 40%. In patients with this strict negative baseline profile, the mPFS was only 6.0 weeks (95% C.I., 5.7 to 6.3) (Fig. 6A). A comparison of negative and non-negative baseline profiles showed hazard ratios for disease progression or death that favored the latter [3.39 (1.4–8.17; 95% C.I.) P = 0.007] (Fig. 6A). Patients were also stratified into hyperprogressors according to the immunological criteria defined in this study, with an mPFS of only 6 weeks (5.7–6.3; 95% C.I.). Hazard ratios for disease progression or death favored patients with non-hyperprogressive T cell dynamics [3.89 (1.58–9.58; 95% C.I.) P = 0.002] (Fig. 6B). To discard that the baseline THD profile was a prognostic instead of a predictive factor, the median time elapsed from diagnosis to enrolment for immunotherapy was evaluated in patients with negative or positive THD profiles. There were no significant differences between the two groups, confirming the predictive and not the prognostic value of THD profiling (10.9 months vs 9.8 months, P=0.9899) (Fig. 6C). ROC analysis was performed in our patient cohort to test the robustness of THD quantification prior to therapy as a predictive biomarker (Fig. 6D). The association of baseline THD cells with clinical output was very significant (R=0.84, P=0.006), with a cut-off value of <57.7% to achieve 100% specificity and 75% sensitivity.
Identification of responders by baseline THD subset profiling and PD-L1 tumor status
Identification of responders with a high probability prior to therapy is currently a major challenge. In this study, all responders belonged to G1 patients and presented a specific THD fingerprint consisting of higher numbers of naïve/stem memory CD4 THD cells compared to G1 non-responders (P=0.03) (Fig. 6E). There was a very significant association between G1 patients with naïve/stem memory THD cells above 5% and objective responses (P=0.003).
PD-L1 tumor expression levels could be evaluated in 21 of the patients before therapy, and did not significantly correlate with baseline THD G1 or G2 profiles (P=0.1). PD-L1 expression correlated with objective responses at the limit of statistical significance in our cohort study when used as a single stratifying factor (P=0.052 by Fishers exact test) (data not shown). PD-L1 tumor expression seemed to segregate G1 patients into responders and non-responders (Fig. 6F), in our limited cohort of patients in whom PD-L1 tumor expression levels could be evaluated. There was a tendency for G1 responder patients compared to G1 non-responders to have higher PD-L1 tumor expression prior to therapy. These results suggested that patients with baseline naïve/stem memory THD cells above 5% together with PD-L1 tumor positivity may accurately identify responders amongst the G1 patient population.
Discussion
Our data shows that the efficacy of PD-L1/PD-1 blockade monotherapies in metastasic NSCLC patients heavily relies on systemic CD4 THD cell numbers. Importantly, we are unequivocally identifying patients in our clinical practice with primary resistance and a high risk of hyperprogressive disease before enrolment, by quantification of THD cells from routine small blood samples. While there has been a recent output of potential biomarkers from blood sampling, most of them have prognostic value rather than predictive capacities, while others are rather challenging to implement in routine clinical practice.
CD8 T cell subsets were extensively studied with similar results as reported (Kamphorst et al., 2017) but without practical stratification capacities. The CD8 T cell response was delayed and followed CD4 THD dynamic changes but at a lesser extent. Therefore, CD8 T cell monitoring had a lack of practical predictive capacities. Initially, we hypothetized that CD4 THD cells in cancer patients were senescent, terminally-differentiated T cells, a subset strongly associated to the EMRA phenotype (Lanna et al., 2017; Lanna et al., 2014). To our surprise, CD4 THD cells in cancer patients were enriched in central and effector memory subtypes, strongly suggesting that THD cells did not reach terminal differentiation. Indeed, CD4 THD cells expressed high levels of Ki67, suggesting that they were proliferating in cancer patients. This raises the question of whether reduced CD28 and CD27 expression in T cells is a true hallmark of T cell senescence, at least in lung cancer patients. Nevertheless, the more differentiated CD4 T cell subsets in lung cancer patients strongly co-expressed PD-1 and LAG3 after stimulation compared to CD28+ CD27+ T cells. This suggests that the systemic pool of differentiated CD4 T cells is dysfunctional in patients, and prone to inactivation by PD-L1/PD-1 interactions. NSCLC patients were separated in two distinct groups according to baseline CD4 THD numbers, strongly suggesting that lung cancer patients had diametrically different immunological responses to cancer. Responder patients presented significantly higher CD4 THD baseline numbers which may represent a pre-existing large pool of antigen-specific central and effector memory T cells with potential anti-tumor capacities. Responders showed decreases in circulating CD4 THD cells following antibody administration, which may indicate either a recovery of CD28-CD27 expression, or mobilization from peripheral blood to secondary lymphoid organs/tumor sites. In contrast, a systemic expansion of central memory THD cells (most likely through active proliferation) was always associated to non-responders which was plainly apparent in hyperprogressors. A negative baseline profile associated to “THD bursts” accurately correlate with hyperprogressive disease in NSCLC without the need of obtaining radiological evidence, which supports hyperprogression as a true phenomenon with a biological basis.
According to our results in our cohort study, NSCLC patients with a negative baseline profile (G2 patients) do not respond to PD-L1/PD-1 immune checkpoint blockade monotherapy. Their identification prior to therapy helps avoiding the enrolment of hyperprogressors and intrinsic non-responders. Of the remaining patients, quantification of CD4 THD cells with naïve/stem memory phenotypes coupled to tumor expression of PD-L1 can help identifying responders with accuracy. Nevertheless, administration of therapy would not be indicated if THD bursts are observed following the first cycle of therapy.
THD profiling can be performed in patients some time before enrolment into immunotherapies without necessarily quantifying this subset right before starting therapy. Indeed, our immunological profiling helps to make objective decisions regarding NSCLC patients under PD-L1/PD-1 blockade, and places CD4 T cell responses at the center of anti-tumor immunity.
Materials and Methods
Study Design
The study was approved by the Ethics Committee at the Hospital Complex of Navarre, and strictly followed the principles of the Declaration of Helsinki and Good Clinical Practice guidelines. Written informed consent was obtained from each participant, and samples were collected by the Blood and Tissue Bank of Navarre, Health Department of Navarre, Spain. 27 patients diagnosed with non-squamous and 7 with squamous NSCLC were recruited at the Hospital Complex of Navarre (Table S1 in supplementary material). Patients had all progressed to first line chemotherapy or concurrent chemo-radiotherapy. Eligible patients were 18 years of age or older who agreed to receive immunotherapy targeting PD-1/PD-L1 following the current indications (table S1). Tumor PD-L1 expression could be quantified in 21 of these patients before the start of therapies. Measurable disease was not required. The exclusion criteria consisted on concomitant administration of chemotherapy or previous immunotherapy treatment. NSCLC patients had an age of 64.8 ± 8.3 (mean ± standard deviation, S.D., N=45). Age-matched healthy donors were recruited from whom written informed consent was also obtained, with an age of 68.60 ± 8 (mean ± S.D., N=40).
Therapy with nivolumab, pembrolizumab and atezolizumab was provided following current indications (Herbst et al., 2016; Horn et al., 2017; Rittmeyer et al., 2017). 4 ml peripheral blood samples were obtained prior and during immunotherapy before administration of each cycle. PBMCs were isolated as described (Escors et al., 2008) and T cells analysed by flow cytometry. The participation of each patient concluded when a radiological test confirmed response or progression, with the withdrawal of consent or after death of the patient. Tumor responses were evaluated according to RECIST 1.1 (Eisenhauer et al., 2009) and Immune-Related Response Criteria (Wolchok et al., 2009). Hyperprogression was identified by the criteria proposed by Saada-Bouzid et al. (Champiat et al., 2017; Saada-Bouzid et al., 2017). Objective responses were confirmed by at least one sequential tumor assessment.
Flow cytometry
Surface and intracellular flow cytometry analyses were performed as described (Gato-Canas et al., 2017; Karwacz et al., 2011). T cells were immediately isolated and stained. 4 ml blood samples were collected from each patient, and PBMCs isolated by FICOL gradients right after the blood extraction. PBMCs were washed and cells immediately stained with the indicated antibodies in a final volume of 50 μl for 10 min in ice. Cells were washed twice, resuspended in 100 μl of PBS and analyzed immediately. The following fluorochrome-conjugated antibodies were used at the indicated dilutions: CD4-FITC (clone M-T466, reference 130–080–501, Miltenyi Biotec), CD4-APC-Vio770 (clone M-T466, reference 130–100–455, Miltenyi Biotec), CD4-PECy7 (clone SK3, reference 4129769, BD Biosciences,) CD27-APC (clone M-T271, reference 130–097–922, Miltenyi Biotec), CD27-PE (clone M-T271, reference 130–093–185, Miltenyi Biotec), CD45RA-FITC (reference 130–098–183, Miltenyi Biotec), CD62L-APC (reference 130–099–252, Miltenyi Biotech), CD28-PECy7 (clone CD28.2, reference 302926, Biolegend), PD-1-PE (clone EH12.2H7, reference 339905, Biolegend), CD8-FITC (clone SDK1, reference 344703, Biolegend), CD8-APC-Cy7(clone RFT-8, reference A15448, Molecular probes by Life technologies).
Cell culture
Human lung adenocarcinoma A549 cells were a kind gift of Dr Ruben Pio, and were grown in standard conditions. These cells were modified with a lentivector encoding a single-chain version of a membrane bound anti-OKT3 antibody.
Spanning-tree progression analysis of density-normalized events (SPADE)
Flow cytometry data was analyzed with SPADE v3 (Qiu et al., 2011) using forward, scatter, CD4, CD28 and CD27 expression as overlapping markers for cell clustering. Arcsinh with cofactor of 150 was used for data transformation, from a maximum of 100000 events and 30 clusters.
Data collection and Statistics
T cell percentages were quantified using Flowjo (Lanna et al., 2017; Lanna et al., 2014). The percentage of CD4/CD8 THD (CD28 and CD27 double-negative) and poorly differentiated T cells (CD28+ CD27+) were quantified prior to therapy (baseline), and before administration of each cycle of therapy. Gates in flow cytometry density plots were established taking non-senescent T cells as a reference. Data was recorded by M.Z.I., and separately analyzed thrice by M.Z.I. and H.A.E. independently. Cohen’s kappa coefficient was utilized to test the inter-rater agreement in classification of immunological profiles (κ=0.939).
The mode of action, pharmacokinetics, adverse events and efficacies of the three PD-L1/PD-1 blocking agents are comparable in NSCLC, which act through the interference with the inhibitory interaction between PD-L1 and PD-1 (Herbst et al., 2016; Horn et al., 2017; Rittmeyer et al., 2017). Treatments administered to the patients were allocated strictly on the basis of their current indications, and independently of any variable under study. Hence, in the study design the use of all data was pre-specified to be pooled to enhance statistical power, and thereby reducing type I errors from testing the hypotheses after had hoc subgrouping into specific PD-L1/PD-1 blockers. The number of patients assured statistical power for Fisher’s exact test of 0.95 and superior for Student t and Mann-Whitney tests (G*Power calculator) (Faul et al., 2009), taking into account that the expected proportion of responders using any of the three PD-L1/PD-1 immune checkpoint blockade drugs in NSCLC is around 20% to 25% (if no stratification using PD-L1 tumor expression levels is used) (Herbst et al., 2016; Horn et al., 2017; Rittmeyer et al., 2017). Our study confirmed the therapeutic efficacy of these agents by achieving a 20% response rate. Two pre-specified subgroup analyses in the study were contemplated. The first, baseline T cell values; the second, post-first cycle T cell changes from baseline. The study protocol contemplated the correlation of these values with responses using Fisher’s exact test, paired Student t tests/repeated measures ANOVA (if normally distributed) or U of Mann-Whitney/Kruskal-Wallis (if not normally distributed, or data with intrinsic high variability) to be correlated with responses. Two-tailed tests were applied with the indicated exceptions (see below). Importantly, the treatment administered to the patients was allocated independently of their basal immunological profiles, and strictly followed the current indications for the PD-L1/PD-1 inhibitors.
The percentage of T cell subsets in untreated cancer patients was normally distributed (Kolmogorov-Smirnov normality test), but not in age-matched healthy donors. Hence, to compare T cell values between two independent cancer patient groups, two-tailed unpaired Student t tests were used while comparisons between healthy subjects and cancer patients were carried out with the U of Mann-Whitney. The mean age of cancer patients and healthy donors was 64.8 ± 8.3 (S.D.) and 68.60 ± 8 (S.D.), respectively. Percentages of T cell populations in treated patients were not normally distributed, so response groups were compared with either Mann-Whitney (comparisons between two independent groups) or Kruskall-Wallis for multi-comparison tests if required and as indicated in the manuscript. To confirm the increase or decrease in THD cells between baseline and post-first cycle of therapy in either responders or non-responders, one-tailed paired t tests were carried out. To test changes in Ki67 expression in T cells between baseline and post-first cycle of therapy in either responders or non-responders, two-tailed paired t tests were carried out. Fisher’s exact test was used to assess the association between CD4 THD dynamic profiles or the baseline values of THD cells with clinical responses. Post hoc Cohen’s kappa coefficient test was used to test the agreement of radiological versus immunological criteria for the identification of hyperprogressors.
Progression free survival (PFS) was defined as the time from the starting date of therapy to the date of disease progression or the date of death by any cause, whichever occurred first. PFS was censored on the date of the last tumor assessment demonstrating absence of progressive disease in progression-free and alive patients. PFS rates at 12 and 28-weeks was estimated as the proportion of patients who were free-of-disease progression and alive at 12 and 28 weeks after the initiation of immunotherapies. Patients who dropped out for worsening of disease and did not have a 28-week tumor assessment were considered as having progressive disease. Overall response rate (ORR) was the proportion of patients who achieved best overall response of complete or partial responses.
PFS was represented by Kaplan-Meier plots and long-rank tests utilized to compare cohorts. Hazard ratios were estimated by Cox regression models. For PFS analyses, patients were pre-specified to be stratified only on the basis of their basal THD values to avoid increase in type I errors due to multiplicity by subgroup analyses. Receiver operating characteristic (ROC) analysis was performed with baseline THD numbers and response/non-response as a binary output. Statistical tests were performed with GraphPad Prism 5 and SPSS statistical packages.
Funding
This research was supported by Asociación Española Contra el Cáncer (AECC, PROYE16001ESCO); Instituto de Salud Carlos III, Spain (FIS project grant PI17/02119), a “Precipita” Crowdfunding grant (FECYT). D.E. is funded by a Miguel Servet Fellowship (ISC III, CP12/03114, Spain); M.Z.I. is supported by a scholarship from Universidad Pública de Navarra; H.A. is supported by a scholarship from AECC; M.G.C. is supported by a scholarship from the Government of Navarre.
Author contributions
M.Z.I. designed and carried out experiments, collected data, analyzed data. H. A.E. designed and carried out experiments, collected data, analyzed data. G.F.H. recruited patients, collected data, analyzed clinical data. M.G.C. carried out experiments, collected data, analyzed data. B.H.M. recruited patients, collected data, analyzed clinical data. M.M.A. recruited patients, collected data, analyzed clinical data. M.J.L. recruited patients, collected data, analyzed clinical data. A.F. recruited patients, collected data, analyzed clinical data. L.T. recruited patients, collected data, analyzed clinical data. R.V. supervised the clinical staff, recruited patients, analyzed clinical data. G.K. conceived the project, supervised non-clinical researchers, analysed data and wrote the paper. D.E. conceived the project, supervised non-clinical researchers, analysed data and wrote the paper. All authors participated in the writing of the manuscript.
Competing interests
The authors declare no competing interests.
Acknowledgments
We sincerely thank the patients and families that generously agreed to take part in this study. We are thankful to Drs Luis Montuenga and Ruben Pio for their constructive comments and input.