NCR+ILC3 concentrate in human lung cancer and associate with intratumoral lymphoid structures

Tertiary lymphoid structures (TLSs) are a common finding in non-small cell lung cancer (NSCLC) and are predictors of favourable clinical outcome. Here we show that NCR+ innate lymphoid cell (ILC)-3 are present in the lymphoid infiltrate of human NSCLC and are mainly localized at the edge of tumour-associated TLSs. This intra-tumoral lymphocyte subset is endowed with lymphoid tissue-inducing properties and, on activation, produces IL-22, TNF-α, IL-8 and IL-2, and activates endothelial cells. Tumour NCR+ILC3 may interact with both lung tumour cells and tumour-associated fibroblasts, resulting in the release of cytokines primarily on engagement of the NKp44-activating receptor. In patients, NCR+ILC3 are present in significantly higher amounts in stage I/II NSCLC than in more advanced tumour stages and their presence correlate with the density of intratumoral TLSs. Our results indicate that NCR+ILC3 accumulate in human NSCLC tissue and might contribute to the formation of protective tumour-associated TLSs. NCR+type 3 innate lymphoid cells display lymphoid tissue-inducing ability. Here the authors show that these cells are increased in early-stage human lung cancer, respond to cancer cells and associated fibroblasts by producing cytokines and are associated to intratumoural ectopic lymphoid structures.

I nnate lymphoid cells (ILCs) are emerging as a family of cells involved in innate anti-microbial immunity, tissue remodelling, lymphoid organogenesis and inflammation, in particular at barrier surfaces 1,2 . Natural killer (NK) cells, the first ILC discovered and the most widely distributed in different tissues, are regulated through a variety of activating and inhibitory cell surface receptors [3][4][5] . Current data are compatible with the concept that the ligands for activating NK receptors are expressed primarily by 'stressed' cells (including tumour or virus-infected cells). NKp46 (ref. 6), NKp30 (ref. 7) and NKp44 (ref. 8) are activating receptors (collectively named 'natural cytotoxicity receptors' (NCRs)) that play a major role in tumour cell recognition and killing by NK cells 9 . Although NCRs were found to bind some viral glycoproteins, their ligands on tumour cells have only partially been defined 10,11 . B7-H6 and, very recently, a novel isoform of the mixed-lineage leukemia (MLL5) protein have been shown to be expressed on a large panel of tumour cells 12,13 and are recognized by NKp30 and NKp44 receptors, respectively. Although a role of NK cells in anti-tumour immunity has been established, such a role for other ILC populations remains largely unexplored.
Recently, three subsets of ILC have been identified on the basis of their cytokine expression patterns and dependency on transcription factors [14][15][16][17] . Notably, ILC subpopulations produce an array of cytokines that in variety match those of T-helper cell subsets. Besides conventional NK (cNK) cells, an additional ILC population expressing NCRs, but functionally different from NK cells, has been identified in both humans [18][19][20][21][22] and mice [23][24][25] . Unlike NK cells, in humans this NCR þ cell population is dependent on the orphan nuclear receptor (ROR)gt transcription factor and on the interleukin (IL)-7Ra for its development and function. In addition, these cells constitutively express NKp44 but not CD94, lack perforin, granzymes and death receptors, while they produce IL-22 following cell activation 25 . These cells were termed NCR þ ILC3, to identify a subgroup of human ILC3 distinct from lymphoid tissue inducer (LTi) cells 26 . Fate-mapping approaches showed that these cells are developmentally distinct from cNK cells and localize primarily in mucosal tissues 27,28 , in particular in the intestinal tract, where they are involved in regulation of epithelial homeostasis, immunity to extracellular bacteria and wound healing 14,18,29,30 . Notably, post-natal CD56 þ RORgt þ CD127 þ cells from human tonsils have been shown to express high levels of NKp44 and NKp46 (ref. 31), to produce lymphotoxin-a and to exert LT-inducing ability, at least in vitro 19,31 .
In different pathological conditions, T, B and dendritic cells (DCs) may be organized in structures termed tertiary lymphoid structures (TLSs) containing functional T-and B-cell zones. Such lymphoid tissue neogenesis shares molecular determinants with secondary lymphoid organ (SLO) ontogeny, including the release of LTa1b2 and tumour necrosis factor (TNF), and the expression, on stromal cells, of their respective receptors (LTbR and TNF receptor (TNFR)) 32,33 . These molecular interactions play a crucial role in the regulation of lymphocyte homing and compartmentalization, in both SLOs and TLSs, by inducing the expression of chemokines and adhesion molecules 34,35 . In mice, during the early stages of SLO development, membrane-bound LTa1b2 is expressed by fetal CD3 neg CD4 þ CD45 þ LTi cells 36 ; however, no clear evidence has been obtained that the adult counterparts of LTi cells (or ILCs with LTi activity) may orchestrate TLS formation or be recruited into inflamed tissues.
Adult bronchus-associated lymphoid neogenesis is now recognized as a common feature of many chronic inflammatory diseases of the lung, such as tuberculosis 37 and chronic obstructive pulmonary disease 38 , and its formation is apparently independent from LTi cells 39 . However, limited information exists on lymphoid neogenesis in cancer, where it remains to be clarified whether TLS formation may represent a beneficial or, rather, a harmful event during cancer progression. In the case of human non-small cell lung cancer (NSCLC), it has been shown that TLSs are associated with a favourable prognosis, suggesting that they may participate in antitumoral immunity 40 . On the other hand, their formation could also represent a detrimental event, as suggested by a murine model of melanoma 41 . Therefore, it is important to clarify whether ILCs are present in tumours, in particular in humans, to define their functional capabilities and their possible involvement in tertiary lymphoid organ neogenesis.
Here we identified NCR þ ILC3-infiltrating human NSCLC. These cells produced proinflammatory cytokines and chemotactic factors on interaction with tumour cells or tumour-associated (TA) fibroblasts. Remarkably, tumour-infiltrating NCR þ ILC3 accumulated at the edge of TA lymphoid aggregates and were endowed not only with proinflammatory but also with LTi functional capability.

Results
NCR þ ILC3 accumulate in NSCLC tissues. We have previously shown that a CD3 neg CD56 þ NKp44 þ cell population selectively infiltrates NSCLC, whereas it is virtually undetectable in normal lung tissues 42 . In addition, NCR þ cells are mainly localized in the tumour stroma, in particular within stromal regions characterized by high density of lymphocytes 42 . As NCR þ ILC3 express high levels of NKp44, these data prompted us to further investigate whether the various ILC subsets, and in particular Group 3 ILC expressing NCRs, were present in the tumour microenvironment. To this end, we employed a more stringent gating strategy, as shown in Fig. 1a, which allowed us to assess the frequency of the different ILC subsets in NSCLC tissue as well as in matched normal lung tissue and in peripheral blood (PB) of the same patient. To better characterize the presence of NCR þ ILC3, NCR neg ILC3 and ILC1 among CD127 þ ILC, we further analysed NKp44 þ CD117 þ cells for the expression of suitable markers (that is, RORgt, CCR6 and CD25). On the other hand, for the characterization of the different subsets of CD127 neg, ILCs, NKp44, NKp46, CD103 and T-bet were analysed. As expected, all CD127 neg NKp46 þ ILCs expressed the transcription factor T-bet and, interestingly, the majority of NKp46 þ NKp44 þ cells expressed CD103, thus resembling the newly described intraepithelial ILC1 (ref 22 and Fig. 1b).
As a whole, ILC composition in NSCLC tissues deeply differed from that in matched normal tissues and in patient PB (Fig. 1c). Among total Lineage neg CD127 þ lymphoid cells, the frequency of NCR þ ILC3 was significantly higher in tumour tissues (44.3% ± 4.5 of total CD127 þ cells) than in matched normal lung (23%±7.2) or than in PB, where they were virtually absent (1% ± 0.5). No substantial difference in the frequency of NCR neg ILC3 was observed between NSCLC and the normal counterpart or lung cancer patient PB. On the contrary, tumourinfiltrating ILC2 were present in significantly lower amount (7.3% ± 2.6) if compared with normal tissues (20.6% ± 5.9). ILC1, which were mainly represented in patient PB, were present in similar amount in tumour and normal lung tissues. Thus, a specific accumulation of NCR þ ILC3 occurs during the development of NSCLC. The preferential accumulation of NCR þ ILC3 in tumour was more evident when their absolute number per mg of tissue was calculated (22.8 ± 7.7 cells per mg in the tumour versus 2.9±1 cells per mg in the normal lung) (Fig. 1d).
These data demonstrate that human NSCLC are infiltrated by a substantial amount of NCR þ ILC3, otherwise sparsely represented in adjacent healthy lung tissues.
NSCLC NCR þ ILC3 are an abundant innate source of cytokines. A specific feature of NCR þ ILC3 is the production of IL-22 on IL-23 stimulation 18 . We therefore stimulated, NCR þ ILC3 freshly isolated from NSCLC tissues with phorbol myristate acetate (PMA)/ionomycin and/or IL-23, and the production of the distinctive pattern of cytokines (IL-22, TNF, IL-8 and IL-2) was detected by intracytoplamic staining (Fig. 1e and Supplementary  Fig. 1a), thus mirroring cytokines produced on the same stimulation by tonsil NCR þ ILC3. However, a lower amount of IL-22 was released by tumour ILCs in comparison with the tonsillar counterpart ( Supplementary Fig. 1b).
To complete their functional characterization, tumour NCR þ ILC3 were expanded in vitro. As a control, NCR þ ILC3 were also isolated from human tonsils and cultured. Thus, NKp44 þ CD127 þ CD117 þ cells were purified by flow cytometry cell sorting and cultured for 4-6 weeks in the presence of IL-2 and IL-7. On in-vitro proliferation, cells maintained all the hallmarks of NCR þ ILC3. When compared with cultured cNK cells, they expressed high levels of CD117, CCR6, CXCR5 and CD25, and higher levels of NKp44 and CD161. In addition, they stably expressed RORgt, while surface NKp30 and NKp46 were expressed at lower levels than cNK cells. On the other hand, tumour-infiltrating NCR þ ILC3 lacked typical cNK cell markers such as CD94, KIR and Perforin (Fig. 2a).
On stimulation, the cytokine expression profile was comparable to that of tonsil NCR þ ILC3, while it was markedly distinct from that of cNK cells (Fig. 2b,c). Both tumour and tonsil NCR þ ILC3 produced significant amount of IL-22, TNF-a, IL-8 and IL-2 on stimulation. On the contrary, differently from cNK cells, IFNg production could barely be detected, whereas IL-17 was not produced by any of the ILC subsets analysed (Fig. 2b,c). To quantify the amount of cytokines released by tumour NCR þ ILC3, the production of relevant cytokines was also confirmed by assessing their presence in culture supernatants on stimulation by PMA/ionomycin and IL-23 (Fig. 2d).
In addition, tumour-infiltrating NCR þ ILC3 did not produce IL-13, IL-5, IL-4 and IL-10 (data not shown). It is noteworthy that the same cytokine pattern was detectable in NCR þ ILC3 freshly isolated from cancer specimens (Fig. 1e), thus confirming that tumour-infiltrating NCR þ ILC3 maintain a stable phenotype after their ex-vivo culture and expansion in the presence of IL-7 and IL-2.
Taken together, these findings suggest that NCR þ ILC3 may play a role in tumour microenvironment because of the release of proinflammatory cytokines (TNF-a and IL-22) and by providing an innate source of cytokines mediating leukocyte recruitment and growth, such as IL-8 and IL-2, respectively.     Fig. 2). First, we assessed whether NCR þ ILC3 could exert any cytotoxic activity against lung tumour cells. No substantial cytolytic activity against NSCLC cells could be detected even at high effector/target (E/T) ratios, as predictable by the absence of lytic granules (Fig. 2a). Conversely, cNK cells efficiently lysed the same target cells (Fig. 3a). Next, we assessed whether TA NCR þ ILC3 could inhibit tumour cell growth in co-culture experiments. Also in this case, virtually no inhibition could be detected in a 4-day proliferation assay. Only a marginal inhibition was observed at 50:1 ratio (Fig. 3b).
Although NCR þ ILC3 did not show any cytolytic and tumourgrowth suppressing abilities, they were able to efficiently recognize lung tumour cells as revealed by cytokine production after co-culture with a variety of lung tumour cell lines (A549, NCI-H1975 and SK-MES1) (Fig. 3c). Recognition of NSCLC cells by tumour NCR þ ILC3 resulted in a strong production of TNF and IL-8 but very low level of IL-2 and IL-22. The production of IL-17 was never detectable. As control, epithelial cells from normal lung (isolated as described in Methods section) were also employed as target cells and, as expected, no cytokine production by tumour NCR þ ILC3 was observed.
It has recently been reported that the engagement of NKp44 in RORgt þ ILC can trigger a coordinated proinflammatory programme 30 . Therefore, we investigated whether NCRs could be involved in the cytokine production induced by the interaction between NCR þ ILC3 and lung tumour cells. Thus, cytokine production was analysed on NCR þ ILC3 exposure to lung tumour cells in the presence or in the absence of blocking monoclonal antibodies specific for NKp44, NKp46 and NKp30. In line with the differences in the levels of expression of NKp30, NKp46 and NKp44 (Fig. 2a), NKp44 receptor appeared as the most involved NCR during interactions with the different cancer cell lines. Remarkably, blocking of NKp44 receptor caused a marked and consistent decrease in the production of TNF-a and IL-8, whereas blocking of other NCRs resulted in a marginal inhibition of cytokine production (Fig. 3d). These data were confirmed by also assessing the amount of cytokines released in the culture supernatants (Fig. 3e). Interestingly, we confirmed the effect of blocking NKp44 also on freshly isolated NCR þ ILC3, as both IL-8 and TNF production was significantly inhibited by blocking the NKp44 receptor ( Fig. 3f-h).
Despite stimulation of tumour NCR þ ILC3 by NSCLC cells resulted in low production of IL-2 and IL-22, we observed that the recognition of cancer cells by NKp44 strongly increased the production of IL-2 and IL-22 induced by the proinflammatory cytokines IL-1b and IL-23 ( Fig. 3i-k).
These data clearly indicate that tumour NCR þ ILC3 can efficiently recognize cancer cells, and that NKp44 plays a relevant role in this interaction. Of note, the strength of NKp44-mediated triggering of NCR þ ILC3 was not due to a higher expression of the NKp44 ligand(s) on NSCLC cells employed as cell target. Indeed, the three NCR ligands were similarly involved in the recognition of NSCLC cells by cNK cells, with NKp46 ligand actually playing a major role ( Supplementary Fig. 2), thus ruling out a preferential expression of NKp44 ligand(s) on target cells.
NCR þ ILC3 produce IL-8 upon interaction with TA fibroblasts. Fibroblasts are an important component of the tumour microenvironment. They have recently been reported to exert a strong immunosuppressive activity on cNK cells 44 . Therefore, we further analysed the effect of the interactions between NCR þ ILC3 and lung TA fibroblasts. Remarkably, NCR þ ILC3 efficiently recognized TA fibroblasts, as this interaction led to the production of TNF-a and IL-8 but not of IL-22 (Fig. 4a). Interestingly, the production of IL-8 was consistently higher than TNF-a, observed as both intracytoplasmic staining (Fig. 4a) and cytokine content in the culture supernatants (Fig. 4d).
As we noticed that TA fibroblasts expressed the ligands for NKp44 and, partially, NKp30 (Fig. 4b), we analysed whether these NCRs were involved in the interaction between tumour-infiltrating NCR þ ILC3 and TA fibroblasts. Antibodymediated blocking of NKp44 strongly inhibited the production and the release of TNF-a but not IL-8 (Fig. 4c,d). On the contrary, blocking of NKp30 and NKp46 had no inhibitory effect.
Overall, these results indicate that NCR þ ILC3 might actively interact with fibroblasts present in the tumour microenvironment, leading to the release of high amounts of leukocyte chemotactic factors. At the same time, they release proinflammatory cytokines such as TNF-a influencing the vascular permeability and acting on other neighbouring cell types of both haematopoietic and non-haematopoietic origin.
Tumor NCR þ ILC3 display LTi potential and associate with TLS. The prototypical group 3 ILC are LTi cells, which are crucial for the formation of SLOs during embryogenesis. Although this ILC subset lacks NCRs and can distinctively produce IL-17A, it shares many development and functional features with NCR þ ILC3, which justify their joint subgrouping as ILC3.
Therefore, we hypothesized that NCR þ ILC3 isolated from NSCLC might show LT-inducing properties. It has previously been reported that Lin neg NKp44 þ CD127 þ cells (both CD56 þ and CD56 neg ) induce the expression of vascular cell adhesion molecule-1 (VCAM-1) and intercellular adhesion molecule 1 (ICAM-1) on mesenchymal stem cells (MSCs) when co-cultured for 4 days in the presence of IL-7. This functional property has been regarded as a marker of the LT-inducing ability mainly due to the production of specific molecules such as TNFa and lymphotoxins (LTa1b2 or LTa3) 19,31 . Therefore, we assessed whether tumour-infiltrating NCR þ ILC3 had this property as well. NCR þ ILC3 isolated from tonsils were used as a positive control. As shown in Fig. 5a, lung cancer-derived NCR þ ILC3 induced a significant upregulation of the adhesion molecules ICAM-1 and VCAM-1 on MSCs.
The expression of lymphotoxins a and b is considered another relevant LTi property that has previously been detected in human tonsil NCR þ ILC3 (ref. 19). We found that NCR þ ILC3 isolated from lung cancers expressed lymphotoxins messenger RNA at levels comparable to those of the tonsil counterpart (Fig. 5b). Because of these latter data and of evidence of an abundant TNFa production by tumour NCR þ ILC3, we investigated the role of these factors in the upregulation of adhesion molecules on MSCs. LTa1b2 binds LTb receptor (LTbR) on MSCs, whereas TNFa and LTa3 share the TNFRII. Thus, by preventing the engagement of these factors on their receptors using TNFRII-and LTbR-soluble molecules, individually or in combination, we assessed their role in the upregulation of adhesion molecules on MSCs. Both pathways appeared involved and the synergistic effect of the two signals seems required for the upregulation of VCAM-1 (Fig.5c).
In line with these findings on possible LTi properties of intratumoral NCR þ ILC3, we observed that these cells were localized at the edge of lymphoid structures associated to NSCLC. Double staining of lung tissue sections with anti-NKp46 and anti-RORgt Abs revealed that in the normal lung tissue, in the peri-bronchial connective tissue, below the normal respiratory epithelium, NKp46 þ RORgt þ lymphoid cells were undetectable. In contrast, in lung adenocarcinomas, in which lymphoid-like follicles are usually detectable (Fig. 5d (i)), NKp46 þ RORgt þ cells were interspersed within the stromal compartment and were often localized at the edge of lymphoid aggregates (Fig. 5d (ii,iii)) or near small blood vessels (Fig. 5d (iv)) On the basis of their functional properties and of their location within tumour tissue, we analysed whether the presence of NSCLC-associated NCR þ ILC3 could be related to the formation of ectopic lymphoid tissues. We found that both the percentage of NCR þ ILC3 in total leukocytes (CD45 þ cells) infiltrating the tumour and the absolute number of tumour NCR þ ILC3 per mg of NSCLC tissue correlated with the density of TLSs in cancer tissues (Fig. 5e). This correlation was independent from the amount of leukocyte infiltrating the tumour, as the absolute number of tumour-infiltrating CD45 þ cells did not correlate with TLSs.
These results suggest that NCR þ ILC3-infiltrating lung cancers are endowed with LTi properties and might contribute to the formation and/or maintenance of TLSs or lymphoid aggregates at the tumour site.
ILC3 triggering via NCRs results in endothelium activation. The recruitment of leukocytes from blood into peripheral tissues requires the passage of cells across the vascular endothelium. This process is tightly controlled and regulated by the expression of adhesion molecules on endothelial surfaces. As NCR þ ILC3 produce large amounts of pro-inflammatory (TNF-a) and chemotactic/angiogenic factors (IL-8), we assessed the activating potential of these factors on endothelial cells.
NCR þ ILC3 were triggered using a mixture of anti-NKp44, NKp30 and NKp46 monoclonal antibodies and, after 24 h, the supernatants were collected and employed for functional assays on endothelial cells. Supernatants derived from NCR-triggered NCR þ ILC3 acted on endothelial cells by strongly increasing their adhesive properties on gelatin-coated wells, as observed by a celladhesion test (Fig. 6a). The increased adherence of endothelial cells to extracellular matrix was accompanied by the upregulation of adhesion molecules involved in cell-to-cell adhesion, including a statistically significant upregulation of ICAM-1 and VCAM-1 (Fig. 6b), most probably secondary to TNF release. These results indicated that an innate activation of NCR þ ILC3 might favour the recruitment of other leukocytes in tumour tissue. As NCR þ ILC3 isolated from NSCLC produced high amount of IL-8, also known as neutrophil chemotactic factor, we investigated, whenever possible, whether the presence of intratumoral NCR þ ILC3 might influence the recruitment of neutrophils. Although many other cells in the tumour microenvironment may represent a suitable source of IL-8, including macrophages, epithelial and endothelial cells, we found an interesting correlation between the amount of neutrophils and of NCR þ ILC3-infiltrating NSCLC tissues ( Supplementary Fig. 3b).
On the other hand, IL-8 released by tumour NCR þ ILC3 does not seem to be involved in angiogenetic processes. Supernatants derived from NCR-triggered NCR þ ILC3 did not modify the basal migration of endothelial cells in vitro (Fig. 6c), thus suggesting that neo-angiogenesis would not represent a main consequence of TA NCR þ ILC3 activation.
Collectively, these findings indicate that NCR þ ILC3 present in the lung tumour microenvironment may contribute to promote an inflammatory activation of the endothelium (see also their detection in close proximity of intratumoral blood vessels, Fig. 5d (iv)). This would favour the recruitment of PB leukocytes and their infiltration and accumulation within the tumour mass.
Frequency of tumour NCR þ ILC3 decreases in advanced NSCLC. As we found a correlation between the density of intratumoral TLS in NSCLC and the amount of NCR þ ILC3 in the tumour infiltrate, and it has been reported that these TLSs correlate with higher survival rate 40 , we hypothesized that NCR þ ILC3 might also be associated with a better tumour prognosis. A major prognostic factor for the survival of lung cancer patients is tumour stage 45,46 . Analyses of possible correlation between NCR þ ILC3-infiltrating lung tumours and tumour stages revealed that samples from patients in stage I and stage II contained proportions of NCR þ ILC3 significantly higher than those from patients in stage III (mean value 0.4%, and 0.41% versus 0.14% of CD45 þ cells, respectively) (Fig. 7). No significant differences in the amount of NCR þ ILC3 were observed between cancer specimens of patients affected by adenocarcinomas or squamous cell carcinomas. In addition, smoking habit did not seem to influence NCR þ ILC3 accumulation in cancer tissues.
Thus, NCR þ ILC3 are present in the microenvironment of NSCLC but are prone to disappear on disease progression. A low amount of NCR þ ILC3 within NSCLC appears indeed associated with more advanced tumour stages and, therefore, potentially, with a poorer prognosis.

Discussion
Lung cancer is the most common cause of cancer-related mortality in the world. Over 80% of cases are represented by NSCLCs, which include adenocarcinomas and squamous cell carcinomas. In such tumours, the presence of TLS composed of mature DC/T-cell clusters and B-cell follicles has been widely documented 47 . Such structures have not been detected in sites distant from the tumour lesions, suggesting that they are likely to be formed in response to tumour cells and their microenvironment. Remarkably, it has recently been shown through univariate analysis that the density of TLS is predictive of survival in patients with early-stage NSCLC 40 .
Our present results show that NCR þ ILC3, a recently described subset of human innate lymphocytes, are present in lung tumours and may be involved in the production of a proinflammatory response in the tumour microenvironment. Analysis of their ARTICLE functional properties revealed that NCR þ ILC3 may promote the recruitment of other leukocytes to the tumour site by secreting chemotactic factors and activating TA vessels. Of note, tumour NCR þ ILC3 also display peculiar LTi properties and might therefore contribute to the formation and/or maintenance of tertiary lymphoid organs or lymphoid aggregates at the tumour site. In agreement with this hypothesis we found that the amount of NCR þ ILC3 within tumour infiltrate correlated with the density of intratumoral TLS. Although other lymphocytes of the tumour microenvironment may produce a similar pattern of soluble factors, including LTa/b and TNF-a, which are critical for the possible LTi properties, NCR þ ILC3 may play an exclusive role in the tumour microenvironment: first, they might be directly activated by cancer cells via NKp44 and therefore represent an early innate source of relevant cytokines; then compared with intratumoral NK cells, which may also be directly activated by cancer cells 42,48 , NCR þ ILC3 apparently lack inhibitory receptors controlling their activities and can also provide a unique innate source of IL-2, a crucial growth factor for the clonal expansion of tumour-specific lymphocytes. On the other hand, the role of an innate source of IL-22 within tumour microenvironment remains to be identified and might even favour cancer cell growth 49 . Nevertheless, it is noteworthy that tumour NCR þ ILC3 consistently released lower amount of IL-22 if compared with tonsillar counterpart (Supplementary Fig. 1b). Other relevant cytokines such as IL-8, TNFs, granulocyte-macrophage colony-stimulating factor 30 could also be released by NCR þ ILC3 on the engagement of NKp44 by cancer cells, again providing critical factors for organizing the immune response. These latter soluble factors can also be released by other cells of the innate immunity, which, however, generally fail to be directly triggered by tumour cells.
Tumour staging is the dominant prognostic factor for NSCLC patient survival 45,46 . Our data show that a higher frequency of NCR þ ILC3 in NSCLC was associated with less advanced stages of NSCLC. Thus, the presence of NCR þ ILC3, as for TLS 40,50 , could be related to a protective immunity and a more favourable tumour prognosis. Further studies, beyond the scope of the present contribution, are clearly required to better document this association and its putative prognostic value.
The evidence of a reduced frequency of NCR þ ILC3 in tumour samples of stage III patients raises the question of the reason of the disappearance of this lymphocyte subset. We could speculate that a more tolerogenic microenvironment is likely to operate in more advanced cancers, where the sustained release of inhibitory cytokines affect most immune cells. The maintenance and proliferation of ILC are promoted by common g-chain cytokines such as IL-2, IL-7 and IL-15, which are mainly produced by haematopoietic cells (mainly myeloid and T cells) and, in part, by non-haematopoietic cells (stromal, endothelial and epithelial cells). Persistent inhibitory environmental signals within tumour tissue could therefore affect the presence of ILC via different pathways acting on different cells present in the tumour microenvironment. However, an active plasticity of ILC has lately been reported 51 and we favour the intriguing hypothesis that a more pronounced inflammation, occurring in later phases of tumour progression, could induce ILC3 to ILC1 transition. However, as the mere count of ILC1 versus ILC3 in tumour tissue would not obviously be sufficient to clarify this issue, further investigation will be required to corroborate this hypothesis.
A tumour is composed of different cell types including malignant, stromal, endothelial and immune cells that form a heterogeneous network and exhibit complex interactions. The finding that ILC expressing NCRs may be present in the tumour tissues is particularly interesting. NCR ligands are highly expressed in tumours as revealed by the direct or indirect (that is, by the use of soluble NCR molecules) detection and by receptor masking-based cytolytic assays 9 . It has recently been reported that RORgt þ ILC can sense and react to environmental signals via NCRs (independently of proinflammatory cytokines) 30 . Here we provide clear evidence that NCR þ ILC3 present in the tumour tissues can be directly triggered by tumour cells, as well as by TA fibroblasts. This occurs mainly via the engagement of the NKp44 receptor, which is highly expressed by NCR þ ILC3-infiltrating NSCLC. We observed an elevated production of both TNF-a and IL-8 on triggering of NKp44 by NSCLC cells. At the same time, as previously reported for tonsil and intestinal ILC3 (ref. 30), we observed a strong synergism for the secretion of IL-22 and IL-2 when TA NCR þ ILC3 where activated by both NKp44 triggering and cytokines such as IL-1b and IL-23. On the other hand, in the case of the interaction with TA fibroblasts, blocking of NKp44 strongly inhibited TNF-a but not IL-8 production. These data suggest that in addition to NKp44, other cellular signals, still unidentified, may act on NCR þ ILC3 for the fine-tuning of their cytokine production.
Thus, activation of NCR þ ILC3 is likely to occur within lung cancer tissues because of the abundant presence of NCR ligands on cells of the tumour microenvironment. The resulting cytokine release may induce leukocyte recruitment and proliferation (IL-8 and a unique innate source of human IL-2) 52,53 , inflammation (TNF-a) 54 and generation/maintenance of lymphoid tissues (LTa/b, TNF) 19 .
Although it is still unclear whether intratumoral TLS represent simple lymphoid aggregates or functional immune structures, their association with a better survival probability suggests that the immune response to the tumour may be initiated in the tumour itself. In this context, it has been reported that extranodal lymphoid structures may participate in the induction of specific immune responses 55,56 . Considering that TLS contain high number of antigen-presenting DCs 40 , the possible occurrence of an intratumoral cancer-specific T-cell priming (in addition to the conventional priming occurring in SLO) could influence the evolution of the tumour and therefore the clinical outcome. tumour-associated antigens (TAAs) should be continuously sampled and stimulate specific T cells, increasing the efficiency and specificity of their priming. During tumour progression, TLS may allow T cells to react more rapidly to the shift in the expression profile of tumour antigens 40 , possibly preventing also peripheral mechanisms of immune tolerance, such as the broad presentation of TAAs by non-professional antigen presenting cells (APC). This hypothesis is in agreement with the better clinical outcome of patients with tumours displaying a higher density of TLS. However, as this protective effect was necessarily established after resection of TLS-containing tumours, these structures may not be directly involved in survival. It is rather conceivable that this alternative priming of adaptive responses may result in the induction of a more effective cancer-specific lymphocyte memory able to better control dissemination of micrometastases on surgical cancer resection.
In accordance with our current results, in an experimental model of melanoma, tumour rejection was promoted by cells expressing NKp46 and depending on the transcription factor RORgt 57 .
It should be mentioned, however, that activation of NCR þ ILC3 within tumours might not always represent a favourable event: a negative effect in colorectal cancer has recently been proposed for IL-22, which would act on epithelial cells and induce Stat3 phosphorylation and cell proliferation 49 . In this regard, it might be argued that, as mentioned above, lung tumour NCR þ ILC3 produce lower levels of IL-22 as compared with tonsillar and, probably, colon-associated NCR þ ILC3. On the other hand, the expression of IL-22 receptors such as IL-22R1 and IL-10R2 can be upregulated by TNF-a, suggesting that IL-22 effects may be amplified by this cytokine 58 , abundantly produced by lung tumour NCR þ ILC3 on NKp44 engagement.
However, on the basis of the association between the amount of tumour NCR þ ILC3 and a more favourable prognosis (that is, their correlation with both TLS density and early stages of disease), we favour the idea of a protective role of this ILC subset in anti-tumour immune response against lung cancer cells.
Altogether, our data revealed that NCR þ ILC3 are an important cellular component of the lymphocytic infiltrate of lung cancers, where they can directly sense tumour and TA cells present in the cancerous microenvironment, mainly via the NKp44-activating receptor. Cell triggering via NKp44 and/or cytokine receptors may result in a relevant innate source of cytokines involved in leukocyte recruitment, endothelial cell activation and lymphoid organ neogenesis. Both NCR þ ILC3 and NKp44 receptor could therefore represent interesting novel therapeutic targets for modulating the immune response against NSCLC. In addition, the identification of NCR þ ILC3 in tumour specimens should also be further investigated as a new immunoscore factor possibly helpful in determining the prognosis of patients operated on for lung cancer.

Methods
Patient selection and samples. Samples were obtained from 57 untreated patients, without any other concomitant lung disease, who underwent surgical resection of NSCLC. Of these, 18 patients were in tumour stage I, 19 in stage II and 20 in stage III. The samples were provided by the Biological Resource Center of IRCCS AOU San Martino-IST-Istituto Nazionale per la Ricerca sul Cancro and the study was approved by the Institutional Ethics Committee of the same Institution. Tonsils were obtained from patients undergoing tonsillectomy at the Giannina Gaslini Institute. The relevant Institutional Review Boards approved the study and all patients gave their written informed consent according to the Declaration of Helsinki.
Cell isolation and sorting. Lung tissue specimens (neoplastic tissue and matched normal lung tissue) were obtained immediately after surgical resection of primary tumours. Both normal and tumour tissues were extensively washed in PBS to remove cell debris and red blood cell aggregates, then the weight was assessed. Samples were then mechanically minced by scissors to obtain small fragments that were enzymatically digested with a cocktail containing DNAse (100 mg ml À 1 , Roche), collagenase (1 mg ml À 1 , Worthington) and hyaluronidase (1 mg ml À 1 , Sigma) in RPMI 1640 supplemented with Pen/Strep for 1.5 h at 37°C. The suspension was then filtered through a cell strainer and, subsequently, washed by centrifugation in PBS to remove residual enzymes. Whenever possible, an aliquot of the cell suspension was used to evaluate the content of polymorphonuclear leukocytes (PMNs), by flow cytometry (FC) (Supplementary Fig. 3). The tissue-cell suspensions from all the samples were then isolated by Ficoll-Hypaque (Sigma) density gradient centrifugation, to obtain tissue-resident mononuclear cells. Human tonsils were mechanically minced by scissors, then filtered through a cell strainer and, subsequently, the tonsil-mononuclear cells were isolated by Ficoll-Hypaque (Sigma). Whenever possible, matched patient's blood samples were also collected and mononuclear cells were isolated by Ficoll-Hypaque (Sigma). Expansion of NCR þ ILC3 and cNK cell lines. Flow cytometry-sorted NCR þ ILC3 and cNK cells were stimulated with irradiated allogeneic peripheral blood mononuclear cells and irradiated 721.221 tumour cells in the presence of PHA (1 mg ml À 1 , Life Technologies), IL-2 (100 U ml À 1 , proleukin; Chiron) and IL-7 (10 ng ml À 1 , Miltenyi). Cytokines were replenished every 2-3 days. Cells were cultured in RPMI 1640 medium (Lonza) plus 1% human AB serum.
Cell cultures. TA fibroblasts were derived from tumour lung specimens. Briefly, cell suspensions derived from tumour specimens were seeded in six-well tissue plates and cultured in RPMI 1640 medium plus 10% FCS. Wells giving rise to cell populations homogeneously displaying fibroblast cell morphology were selected. Fibroblast cell cultures were further phenotypically characterized and assessed for purity by the analysis of anti-fibroblast antibody (Abcam, ab28244, dil: 1:10). Fibroblasts cell lines were cultured in RPMI medium supplemented with FCS 10% and then used within 15 passages. All the human NSCLC cell lines used in the study (A549, NCI-H1975 and SK-MES1) were purchased from Interlab Cell Line Collection of IRCCS AOU San Martino-IST-Istituto Nazionale per la Ricerca sul Cancro and confirmed mycoplasma free.
MSCs and endothelial cell assays. Human MSCs were derived from the bone marrow, as previously described 59 . Samples were obtained after the ethical committee approval of the institutional review board of the Giannina Gaslini Institute, Genova, Italy, and parental informed consent in accordance with the Declaration of Helsinki. Briefly, bone marrow cell suspensions were plated at a concentration of 1 Â 10 6 cells per ml in 25-cm 2 tissue-culture flasks in Mesencult basal medium supplemented with MSC stimulatory supplements (both from StemCell Technologies). After 24 h, non-adherent cells were removed and fresh medium was added. When the cultures nearly reached confluence, cells were detached by treatment with trypsin/EDTA solution (BioWhittaker, Cambrex) and replated at 5 Â 10 5 cells per 75-cm 2 tissue-culture flask.
Double immunohistochemistry and quantification of TLSs. For NKp46/RORgt double staining, paraffin-embedded lung tissue sections were deparaffinized, treated with H 2 O 2 /3% for 5 min to inhibit endogenous peroxidase and then washed in H 2 O. They were then incubated for 30 min with the first primary antibody (mouse anti-NKp46/NCR1 Ab, R&D System, 195314, conc: 5 mg ml À 1 ) followed by detection with the Bond Polymer Refine Detection Kit (Leica Biosystems) according to the manufacturer's protocol and subsequently incubated for 30 min with the second primary antibody (mouse anti-RORgt Ab, Millipore, MABF81, dil: 1:50) followed by detection with the Bond Polymer Refine Red Detection Kit (Leica Biosystems) according to the manufacturer's protocol.
For the quantification of intratumoral TLSs, for each paraffin-embedded lung tumour, two observers (two expert pathologists trained to identify the pathologic features of NSCLC) selected the tumour section containing a representative area of tumour with adjacent lung parenchyma and the highest density of immune cells on the haematoxylin and eosin-safran-stained tissue section. Intratumoral lymphoid structures were counted in the tumoral areas of the entire tissue section (from 18 to 83 fields, original magnification Â 100) and expressed as density of TLS (total number of TLS per field). The necrosis and fibrosis were counted as the percentage of the positive areas among the whole tumour mass.