Antibody-based Delivery of TNF to the Tumor Neovasculature Potentiates the Therapeutic Activity of a Peptide Anticancer Vaccine

Therapeutic cancer vaccination with peptides represents a promising strategy for active immunotherapy, which aims at inducing robust T-cell effector functions against neoplastic lesions. However, despite considerable efforts, clinical translation of peptide vaccines into efficacious therapies for patients with cancer has been challenging (1, 2). Therapeutic immunization with peptides has been evaluated in patients with different types of malignancies including breast cancer, lung cancer, melanoma, pancreatic cancer, and colorectal cancer (1). A survival advantage was observed with some vaccines in phase II trials, but objective regressions for solid tumor lesions were rarely reported (1). Recent advances in the understanding of cancer immunology and mechanisms of immune suppression pathways in the tumor microenvironment (TME) have resulted in novel strategies for the application of peptide vaccines with promising results in preclinical in vivo studies (3, 4). Furthermore, emerging technologies in the discovery of neoantigens from cancer exome sequences have allowed the development of personalized cancer vaccines, which have shown anticancer activity in preclinical models and in patients (5–7). However, to provoke long-lasting protective antitumor immunity with peptide vaccines, suitable combination modalities (which synergistically enhance immunity against tumors, modulate the TME, or block immunosuppressive mechanisms) are still needed (1–3).

The gp70 envelope protein of murine leukemia virus (MuLV) represents an attractive antigen for the study of vaccination strategies in immunocompetent mouse models of cancer. The virus is endogenous in the genome of most laboratory mouse strains (8), but MuLV proteins are usually not detected in healthy tissue. Interestingly, strikingly high levels of gp70 have been observed in a large variety of widely used mouse cancer cell lines (9, 10). AH1, a peptide derived from the gp70 envelope protein of the MuLV, was first described by Huang and colleagues (11) as the major tumor rejection antigen of the BALB/c-derived CT26 colon carcinoma cell line. AH1 has since been used as a model tumor antigen to investigate CD8⁺ T-cell immunity when using CT26 and other mouse tumor models (12, 13). In addition, a number of reports have described the use of AH1 and AH1-related sequences as anticancer therapeutics in immunocompetent mouse models of cancer (14–16). Immune responses against endogenous retroviral proteins have been a matter of intense investigations and are not restricted solely to mouse tumor models. Indeed, retroviral sequences have been found in the genome of all vertebrate species (17). Consequently, retroelements of the human endogenous retrovirus (HERV)-K family have shown high expression in human cancers and a potent T-cell activity against some retroviral antigens has been observed in patients with cancer (18, 19).

Among the various antibody therapeutics that could be considered as combination partners for anticancer vaccines, we focused our attention on F8-TNF. Various studies have shown that the therapeutic index of TNF can be substantially enhanced by fusion to suitable antibody fragments capable of selective localization to the tumor environment (20–23). In addition, synergistic effects between targeted TNF and immunotherapy have been described in the literature (24, 25). F8-TNF is a noncovalent homotrimeric recombinant protein consisting of the scFv(F8) antibody fragment, sequentially fused to murine TNF (21). The F8 antibody (26) recognizes the alternatively spliced extra-domain A (ED-A) of fibronectin, a target which is virtually undetectable in normal adult tissues (with the exception of the placenta, the endometrium in the proliferative phase, and some vessels of the ovaries), while being strongly expressed in the majority of solid tumors (27), lymphomas (28), and in acute leukemias (29). After intravenous administration, F8-TNF exhibits an efficient and selective homing to the tumor site, as revealed by quantitative biodistribution studies with radiolabeled protein preparations and by ex vivo fluorescence microscopy investigations (10, 21). Within the neoplastic mass, F8-TNF can induce a rapid necrosis, turning the tumor into a black, hemorrhagic mass. The therapeutic action of antibody–TNF fusions can be improved by combination with other cytokine-based therapeutics (30–32) or with certain chemotherapeutic agents (10, 20, 21).

We recently observed that immunocompetent BALB/c mice, bearing subcutaneously grafted WEHI-164 sarcomas, could be cured by the combined action of F8-TNF and doxorubicin (10, 21). Interestingly, cured mice rejected subsequent challenges with WEHI-164 tumors, as well as other BALB/c-derived tumors (e.g., CT26 and C51 colon carcinomas), in a process that was driven by CD8⁺ T cells (10). A comparative MHC class I peptidome analysis revealed that AH1 was the immunodominant tumor rejection antigen in this setting (12).

Encouraged by the anticancer activity of F8-TNF used as single agent and by the role played by AH1-specific CD8⁺ T cells in the tumor rejection process, we decided to investigate combination strategies in immunocompentent mice. Here, we report that peptides derived from the gp70 antigen, displayed only a modest tumor growth inhibition activity, while their action could be substantially enhanced by combination with F8-TNF. A FACS analysis of surface markers (including AH1-specific tetramer reagents) revealed that a large portion of CD8⁺ T cells within the tumor mass were AH1-specific and had an immunosuppressed/exhausted phenotype. The combination treatment with F8-TNF was effective, as it decreased the bulk of the solid tumor mass, leaving a minimal residual disease that could be attacked by the vaccine-boosted immune system.

WEHI-164 fibrosarcoma and CT26 colon carcinoma cells were purchased from the ATCC. Cells were handled according to the manufacturer's protocols and kept in culture for no longer than 2 months. Authentication of the cell lines including check of postfreeze viability, growth properties and morphology, test for Mycoplasma contamination, isoenzyme assay, and sterility test were performed by the cell bank before shipment. Eight-week-old female BALB/c mice were obtained from Charles River. All animal experiments were conducted in accordance with Swiss animal welfare laws and regulations under a project license granted by the Veterinäramt des Kantons Zürich, Switzerland (27/2015).

The F8-TNF immunocytokine was produced as described by Hemmerle and colleagues (21). The AH1 peptide (amino acid sequence: SPSYVYHQF) and the AH1-synthetic long peptide (SLP; amino acid sequence: HSPSYVYHQFERRAKYKREPV) were ordered from Biomatik as TFA salts with a purity >98%. Poly(I:C) was obtained from Sigma-Aldrich as lyophilized potassium salt with buffer salts.

Exponentially growing WEHI-164 or CT26 tumor cells were harvested, repeatedly washed, and resuspended in saline prior to injection. Cells were implanted subcutaneously in the right flank of the mice using 3 × 10⁶ cells per animal. Tumor volume was calculated as follows: [length (mm) × width (mm) × width (mm)]/2. When tumors were clearly palpable, mice were randomly divided into the different treatment groups (n = 5). Vaccination was started when tumor sizes reached about 50 mm³. Mice were vaccinated subcutaneously with 50 μg of the peptide in combination with 100-μg poly(I:C) every 4 days. For the AH1 monotherapy study in WEHI-164–bearing mice, control mice were immunized subcutaneously either with saline alone or with poly(I:C) alone. F8-TNF was injected intravenously in the lateral tail vein. Mice received 3 injections of either 1-μg F8-TNF (WEHI-164 tumor-bearing mice) or 2.5 μg (CT26 tumor-bearing mice) every 48 hours starting on the day after vaccination. For combination studies of the peptide vaccines with F8-TNF, mice were grouped as follows: control group [received 100-μg poly(I:C) subcutaneously and saline intravenously], AH1 vaccine group (peptide vaccine subcutaneously and saline intravenously), F8-TNF group [100-μg poly(I:C) subcutaneously and F8-TNF intravenously], and combo group (peptide vaccine subcutaneously and F8-TNF intravenously). Animals were euthanized when tumor volumes reached a maximum of 2,000 mm³ or weight loss exceeded 15%.

Affinity purification of murine MHC I complexes was performed as described previously (10). Lysis of CT26 tumor cells was performed at a density of approximately 2.5 × 10⁷ cells/mL. M1/42 (BioXcell) antibody–coupled resin was used for the purification of MHC I complexes from the cell lysates.

Analysis of MHC I peptides by LC/MS was performed as previously described for human HLA class I peptides by Gloger and colleagues (33). The datasets were searched using the MaxQuant software (34) against all murine proteins (89,527 entries) of the UniProt database, downloaded on the March 22, 2018. MaxQuant parameters were essentially set as described previously (33): (i) digestion mode: unspecific, (ii) first search mass tolerance 20 ppm; main search mass tolerance 4.5 ppm, (iii) fragment mass tolerance 20 ppm, (iv) 1 variable modification: oxidation of methionine, (v) no specific amino acids for the generation of the decoy databases, (vi) peptide FDR 1%, protein FDR 1%, and (vii) peptide length allowed: 8–20 amino acids. Match between runs was allowed with a matching time window of 3 minutes and an alignment time window of 20 minutes. Reverse hits were removed from the “peptide.txt” output file. For comparison reasons, WEHI-164 samples from (10) were reanalyzed with the same parameters. The MaxQuant output tables can be found in Supplementary Table S1.

All 9mers of the CT26 MHC class I peptidome were analyzed by the GibbsCluster-2.0 Server (35) using the default settings without alignment. On the basis of the resulting clusters, the peptides were annotated to the murine MHC class I alleles H2-D^d, -K^d, and -L^d.

CD4 epitope prediction was based on the peptide binding prediction presented previously (36). In brief, the gp70 protein sequence (UniProt accession: Q61919) was cut into all possible 9mers and scored with position-specific scoring matrices from (37). The higher the binding score, the higher the possibility for a given peptide to bind to a given MHC class II complex. All 9mers with a score greater than 6 are presented in Supplementary Table S2. These peptide sequences typically can bind to MHC class II complexes. The 21mer for vaccination was designed to contain the AH1 peptide and 3 of the 4 highest scoring peptide sequences.

Expression of recombinant murine MHC class I H-2L^d heavy chain and of human β2-microglobulin was performed according to established protocols. Refolding of the MHC class I complex with the AH1 peptide was followed by biotinylation with MBP-BirA and size exclusion chromatography (Superdex S75 10/300 GL, GE Healthcare). Assembled monomers were stored in 16% glycerol at −20°C. MHC class I tetramers were produced by addition of APC-conjugated streptavidin (BioLegend) to the monomers at a final molecular ratio of 1:4. H-2L^d tetramers binding the p29 peptide (38) were used as negative control to check binding specificity. Plasmids of the H-2L^d heavy chain and of human β2-microglobulin were a kind gift of Prof. A. Oxenius (ETH Zürich, Switzerland). AH1 and p29 peptides were ordered from Biomatik.

CT26 or WEHI-164 tumor–bearing mice were vaccinated as described above. The injection of F8-TNF was postponed by 4 days, to ensure an adequate tumor size for the analysis of tumor-infiltrating lymphocytes. Tumor-draining lymph nodes (DLNs) and tumors were excised on the day after the first F8-TNF injection. DLNs were minced in PBS, treated with RBC Lysis Buffer (BioLegend), passed through a 40-μm cell-strainer (EASYstrainer, Greiner Bio-One) and repeatedly washed. The single-cell suspension was used directly for flow cytometric analyses. Tumors were cut into small pieces and digested in RPMI1640 (Thermo Fisher Scientific, with l-glutamine) containing antibiotic–antimycotic solution (Thermo Fisher Scientific), 1 mg/mL collagenase II (Thermo Fisher Scientific), and 0.1 mg/mL DNase I (Roche) at 37°C, 5% CO₂ for 2 hours in an orbital shaker set to 110 rpm. The cell suspension was then passed through a 40-μm cell strainer (EASYstrainer, Greiner Bio-One), repeatedly washed and immediately used for flow cytometric analyses.

Approximately 1 × 10⁶ cells from DLNs or 5 × 10⁶ cells from tumors were stained for 30 minutes at room temperature with Zombie Red Fixable Viability Kit (BioLegend) and afterwards incubated on ice for 20 minutes with TruStain fcX (anti-mouse CD16/32, BioLegend). Surface staining was performed with APC-coupled AH1 tetramers and fluorochrome-conjugated antibodies against Thy1.2 (clone 53–2.1), CD8 (53–6.7), CD4 (GK1.5), CD44 (IM7), CD62L (MEL-14), PD-1 (29F.1A12), TIM-3 (RMT3-23), and LAG-3 (C9B7W), which were all purchased from BioLegend. Cells were stained in PBS containing 0.5% BSA and 2 mmol/L EDTA for 1 hour at 4°C. For the staining of intracellular markers, cells were fixed and permeabilized using the eBioscience Foxp3/Transcription Factor Staining Buffer Set according to the manufacturer's protocol. Cells were incubated with fluorochrome-conjugated antibodies against Ki-67 (16A8, BioLegend) and Foxp3 (MF-14, BioLegend) in PBS containing 0.5% BSA and 2 mmol/L EDTA for 30 minutes at room temperature. Cells were analyzed on a CytoFLEX cytometer (Beckman Coulter) and data were processed using FlowJo (v.10, Tree Star).

Data were analyzed using Prism 7.0 (GraphPad Software, Inc.). Statistical significances were determined with a regular 2-way ANOVA test with the Bonferroni posttest. A Student t test was used to assess the differences of the effector-to-tumor cell ratios between the different treatment groups. Data represent mean ± SEM. P < 0.05 was considered statistically significant (*, P < 0.05; **, P < 0.01; ***, P ≤ 0.001; and ****, P < 0.0001).

In a first experiment, we studied the tumor growth inhibition activity of AH1 (amino acid sequence: SPSYVYHQF), following subcutaneous administration to immunocompetent BALB/c mice, bearing established WEHI-164 sarcomas of approximately 50 mm³ in size (Fig. 1A). The peptide, which was given at a dose of 50 μg in combination with 100-μg poly(I:C), did not stabilize the size of the neoplastic mass, but delayed tumor growth significantly more compared with the adjuvant alone or saline (P < 0.0001). When the experiment was repeated, including treatment groups with a suboptimal dose of F8-TNF (1 μg; sarcomas are very sensitive to the action of TNF; refs. 10, 21), a potentiation of anticancer activity was observed, with 3 of 6 cured mice in the combination group (Fig. 1B). Similar investigations were performed in mice bearing subcutaneous CT26 adenocarcinomas. Also in this case, with a full dose of F8-TNF (2.5 μg), cancer cures were observed in the combination group (3/5 mice; Fig. 1D). In an attempt to boost the anticancer activity of the AH1 peptide, we used a SLP of the natural gp70 amino acid sequence, which contains AH1 and 3 putative H2 I-E^d epitopes (amino acid sequence: HSPSYVYHQFERRAKYKREPV; Supplementary Table S2). However, the inclusion of the additional epitope did not improve therapeutic performance (Fig. 1D).

We had previously reported that the AH1 peptide is efficiently presented on MHC class I molecules in WEHI-164 tumors (10). Here, we have investigated if the MHC peptidome of CT26 tumor cells is comparable with the WEHI-164 peptidome. MHC class I peptidome analysis from 5 replicates of 100 million cultured CT26 tumor cells led to the identification of a total of 1,869 sequences (Fig. 2A) with the majority of eluted peptides being 9 amino acids in length (Fig. 2B). A comparison of the WEHI-164 peptidome with the CT26 peptidome showed that 38.5% of all peptide sequences were shared between the 2 cell lines (Fig. 2C). Among the shared peptides was also the AH1 peptide sequence (Supplementary Table S1). A GibbsCluster analysis further revealed that 693, 585, and 552 peptides were eluted from H2-D^d, -K^d, and -L^d alleles (Fig. 2D).

To gain mechanistic insights into the anticancer activity in the various treatment groups, we compared the density of T cells in DLNs and in the tumors of CT26-bearing mice. The highest density of CD8⁺ T cells could be observed in specimens derived from the combination therapy group (P = 0.0182, compared with the control group in DLNs). Moreover, CD8⁺ T cells were more abundant than CD4⁺ T cells within all neoplastic masses (Fig. 3A). A slight elevation in the relative frequency of Foxp3-positive T cells in the tumor was observed in all treatment groups (n = 4 per group), which had received the AH1 peptide (Fig. 3B). 25.2% ± 2.5% of CD4⁺ Foxp3⁺ T cells were detected in tumors of mice treated with the AH1 vaccine alone, which was significantly higher compared with the number in tumors of control mice (15.0% ± 1.9%, P = 0.001). A similar increase was observed with the combination treatment (24.5% ± 1.9%, P = 0.002) whereas the number of regulatory T cells did not change with the F8-TNF treatment alone (15.1% ± 2.6%).

FACS analysis of mice from the various treatment groups (n = 4 per group), performed with a fluorescently labeled AH1-specific tetramer reagent, revealed that AH1-specific CD8⁺ T cells increased in DLNs of mice with tumor (0.55% ± 0.1%), relative to naïve mice (not detectable; Fig. 4A; Supplementary Fig. S1). Treatment with F8-TNF (alone or in combination with AH1 peptide) mediated a further increase in AH1-specific T cells to 1.2% ± 0.2% (P = 0.002) and 1.5% ± 0.4% (P < 0.001), respectively (Fig. 4A). The frequency of AH1-specific CD8⁺ T cells increased substantially within the solid tumor mass, reaching 15.9% ± 1.3% of all T cells in the combination treatment group (Fig. 4B). In this setting, 51.0% ± 4.9% of all CD8⁺ T cells in the tumor were specific to AH1, a proportion that was substantially greater than in mice, which had received only saline plus poly(I:C) (28.9% ± 1.6%, P = 0.001; Fig. 4C).

Furthermore, we observed that most CD4⁺ and CD8⁺ T cells were naïve (CD44^low, ∼80%–90%) in tumor DLNs. However, AH1-specific CD8⁺ T cells had an effector phenotype (CD44^high, >75%) suggesting a possible role in tumor surveillance (Fig. 5A and B). Interestingly, these AH1-specific T cells were positive for the exhaustion markers PD-1, LAG-3, and (to a lesser extent) TIM-3 (Fig. 5C). There were no striking differences among treatment groups, even though mice that had received AH1 (alone or in combination) showed an increased proliferation (Ki-67 staining; Fig. 5C). A similar analysis was performed for tumor-infiltrating T cells. An effector phenotype was observed for both CD4⁺ and CD8⁺ T cells, the latter exhibiting a higher expression of inhibitory surface markers (Fig. 6).

F8-TNF treatment leads to a rapid hemorrhagic necrosis for large portions of the solid tumor mass. This is reflected not only in a FACS analysis of cellular contents of neoplastic masses from the various treatment groups (Fig. 7A), but also in a macroscopic inspection of black scab formation (Fig. 7B), as well as in hematoxylin and eosin staining of tumor sections following treatment (Fig. 7C). In this context, the immune system has to deal with minimal residual disease and it is reasonable to assume that AH1 vaccination is more efficacious when a smaller number of tumor cells needs to be killed. This effect is described by substantially increased effector-to-tumor cell ratios in the combination group (Fig. 7D). A ratio of CD8⁺ AH1⁺ T cells to tumor cells of 0.126 ± 0.026 was observed in the combination group, which was significantly higher than the ratios in the AH1 monotherapy group (0.042 ± 0.020, P = 0.045), the F8-TNF monotherapy group (0.055 ± 0.007, P = 0.039), and the control group (0.015 ± 0.006, P = 0.006).

In this study, we have shown that the AH1 peptide is the main target of CD8⁺ T-cell recognition for 2 types of tumors (CT26 and WEHI-164) grown in BALB/c mice. The use of the AH1 peptide as therapeutic vaccine led only to a modest tumor growth retardation when used as monotherapy, whereas this interventional modality could be boosted by combination with F8-TNF.

The role of AH1 as a tumor rejection antigen in BALB/c mice has been extensively studied (11, 13–16). We have recently reported that mice cured from WEHI-164 sarcomas would reject subsequent challenges with the same tumor cells or with different BALB/c-derived tumor cell lines (10). However, F1F tumors (which are low in gp70 envelope protein of MuLV) were not rejected, thus providing additional evidence for the dominant role of AH1 recognition in tumor surveillance.

The role of immune responses to endogenous retroelements in human malignancies has been a matter of intense investigations. Various studies indicate that human cancer cells reactivate the expression of latent HERV proteins (39, 40). For example, proteins of the HERV-K family have shown abundant expression in various tumors, including melanoma, breast cancer, leukemias, and sarcomas (39). In addition, certain retroviral antigens have been shown to elicit a potent T-cell response in human cancers (19), indicating that their recognition might be an important pathway for host immunity. Although the effects of HERV in cancer remain to be fully elucidated, recent work by different groups indicates their potential use for immunotherapeutic strategies including vaccination or chimeric antigen receptor therapy (39, 41, 42).

The experiments presented in this article suggest that the targeted delivery of TNF may be attractive for the boosting of therapeutic anticancer peptide vaccination strategies. The fusion protein L19-TNF (20), closely related to F8-TNF, is currently being studied in phase III clinical trials for the treatment of metastatic soft-tissue sarcoma, in combination with doxorubicin (Eudra-CT no. 2016-003239-38; refs. 32, 43, 44). While the F8 antibody recognizes the ED-A domain of fibronectin, L19 is specific to the alternatively spliced ED-B domain of fibronectin (45). In mice, the 2 antibodies display a similar tumor-targeting performance. NGR-TNF is a second TNF-based biopharmaceutical, featuring an NGR-containing peptide at the N-terminus of the TNF moiety (46). This peptide recognizes a CD13 aminopeptidase variant, which is often associated with the tumor neovasculature (47). NGR-TNF is currently being tested in a phase III trial in patients with malignant pleural mesothelioma, as well as in randomized phase II trials in 4 different types of solid tumors, either as monotherapy or in combination with chemotherapeutic agents.

AH1-specific CD8⁺ T cells display an immunosuppressed/exhausted phenotype characterized by a high expression of the immune checkpoint markers PD-1, LAG-3, and TIM-3 (Figs. 5 and 6). This surface phenotype did not significantly change between the various treatment groups presented in this study. The chronic exposure of T cells to the AH1 antigen presented on the tumor cells might trigger the increased expression of PD-1, TIM-3, and LAG-3. However, the therapeutic activity of the peptide vaccination and its combination with F8-TNF indicates that AH1-specific CD8⁺ T cells are not totally inert and may exert lytic functions. These results are in line with a study previously published by our group, showing the importance of AH1 recognition for the tumor rejection in F8-TNF–treated mice and their ability to lyse tumor cells (10). In addition, similar phenotypes were observed in neoantigen-specific lymphocytes in the peripheral blood of patients with melanoma, in tumor antigen-specific CD8⁺ T cells induced by melanoma vaccines in patients (48, 49), as well as in tumor-infiltrating T cells for other malignancies (e.g., ref. 50). It is reasonable to assume that a combination with immune checkpoint inhibitors may further boost therapeutic peptide vaccination (1–3). A combination of peptide vaccination with PD-1 blockade and an engineered cytokine (IL2-Fc) has recently been reported to cure different mouse models of cancer (4).

The synergistic effect between F8-TNF and peptide vaccination relied on 2 effects. First, an increased density of AH1-specific CD8⁺ T cells was observed in both the tumor and DLNs after treatment with F8-TNF and after vaccination. Combination of the 2 modalities further increased the amount of the tumor-specific T cells. Second, F8-TNF mediated a potent anticancer effect by triggering a rapid hemorrhagic necrosis within the tumor mass, which led to a significant increase of the effector-to-tumor cell ratio in the combination group. Interestingly, massive hemorrhagic necrosis has also been observed in superficial human cancers after intratumoral and intravenous applications of L19-TNF (32, 44). L19-TNF can be safely administered up to 13 μg/Kg (43), which converts to a mouse equivalent dose of 3 μg (i.e., similar to the 2.5 μg used in this study for the F8 fusion with murine TNF). The advent of perfusion MRI methodologies appears to be ideally suited to assess whether a similar process may also be observed in internal tumors. Demonstration of this pharmacodynamic effect in clinical trials would provide a strong rationale for the combination of TNF-based biopharmaceuticals with anticancer vaccines.

D. Neri has ownership interests (including patents) and is a consultant/advisory board member for Philogen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P. Probst, T. Fugmann, D. Neri

Development of methodology: P. Probst, M. Stringhini, D. Neri

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Probst, M. Stringhini, D. Ritz, T. Fugmann, D. Neri

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Probst, M. Stringhini, D. Ritz, T. Fugmann

Writing, review, and/or revision of the manuscript: P. Probst, D. Neri

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Ritz, D. Neri

Study supervision: D. Neri

Financial support by the ETH Zürich, the Swiss National Science Foundation (grant number 310030B_163479/1), the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement 670603), and the Federal Commission for Technology and Innovation (KTI, grant number 12803.1 VOUCH-LS) is gratefully acknowledged. D. Neri gratefully acknowledges ETH Zürich, the Swiss National Science Foundation (grant number 310030B_163479/1), the European Research Council (ERC, under the European Union's Horizon 2020 research and innovation program, grant agreement 670603), and the Federal Commission for Technology and Innovation (KTI, grant number 12803.1 VOUCH-LS) for financial support.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.