Graphical abstract

   graphic file with name fx1.jpg
   [45]Open in a new tab

Highlights

     * •
       apCAFs are associated with unfavorable outcomes in NSCLC patients
       receiving NCIT
     * •
       apCAFs’ expansion is triggered by IFN-γ via the JAK1/2-STAT1
       pathway
     * •
       apCAFs promote the formation of FOXP1^+Tregs through the PD-L2-RGMB
       axis
     * •
       Targeting apCAFs by the blockade of the PD-L2-RGMB axis improves
       immunotherapy efficacy
     __________________________________________________________________

   Cao et al. reveal that antigen-presenting cancer-associated fibroblasts
   (apCAFs) hinder NCIT effectiveness in NSCLC. IFN-γ induces apCAF
   expansion, which promotes FOXP1^+Treg formation through the PD-L2-RGMB
   axis. Targeting this pathway enhances immunotherapy outcomes, offering
   a potential therapeutic strategy.

Introduction

   Neoadjuvant therapy has revolutionized perioperative regimens for
   cancer patients.[46]^1^,[47]^2^,[48]^3 One notable example is
   neoadjuvant chemoimmunotherapy (NCIT), which was associated with
   significantly greater event-free survival than neoadjuvant chemotherapy
   (NCT) alone.[49]^3 The efficacy of neoadjuvant therapy is primarily
   assessed by the major pathological response (MPR) rate, defined as less
   than 10% residual viable tumor in the resected specimen.[50]^4
   Non-small cell lung cancer (NSCLC) was the first cancer type reported
   to benefit from neoadjuvant immune checkpoint inhibitors
   (neo-ICIs).[51]^5 Meanwhile, NCIT has become the dominant therapeutic
   approach in the neoadjuvant setting, extending its use in clinical
   practice due to its promising clinical response rates compared to
   either NCT[52]^6 or neo-ICIs alone.[53]^7 The NADIM-2 phase 2
   randomized trial highlighted that the pathologically complete
   regression (pCR) rate improved to 37% in the nivolumab plus
   chemotherapy group compared to 7% with chemotherapy alone.[54]^6
   Furthermore, the KEYNOTE-671[55]^6 and AEGEAN[56]^8 trials indicated
   that neoadjuvant anti-PD1/L1 regimens plus chemotherapy confer greater
   benefits compared to NCT alone.[57]^9^,[58]^10 Despite these promising
   results, the non-MPR rate remains high, reaching approximately 60% in
   the CheckMate 816 trial.[59]^11 Furthermore, the LCMC3 trial reported a
   higher non-MPR rate of 80%, with 9% experiencing progressive disease
   (PD). However, the specific mechanisms responsible for non-MPR remain
   largely unknown.[60]^12 This clinical evidence underscores the need to
   develop new strategies to overcome resistance to NCIT.

   Cancer-associated fibroblasts (CAFs) are the predominant type of
   stromal cells in the tumor microenvironment (TME).[61]^13 CAFs can
   induce immunosuppressive effects by interacting with T cells[62]^14 and
   tumor-associated macrophages,[63]^15 among others. Numerous studies
   have demonstrated their tumor-promoting behaviors. Recently, a few
   studies exploring their role in immunotherapy resistance have emerged.
   Liang et al. unveiled that autophagy-enhanced CAFs could compromise
   immunotherapy by up-regulating PD-L1.[64]^14 Similarly, Lin et al.
   identified biglycan-secreting CAFs as a barrier to anti-PD1
   efficacy.[65]^16 However, the role of specific CAF subtypes in inducing
   anti-PD1 resistance in NSCLC, particularly in the context of NCIT,
   remains unexplored. In this study, using single-cell RNA sequencing
   (scRNA-seq), digital spatial profiling (DSP) spatial sequencing, and
   bulk RNA sequencing (RNA-seq) of patients’ tumor samples during NCIT,
   we identified that antigen-presenting cancer-associated fibroblasts
   (apCAFs) are associated with NCIT non-MPR in NSCLC patients.

   apCAFs are a newly identified CAF subcluster characterized by major
   histocompatibility complex (MHC)II expression.[66]^17 Brekken et al.
   reported apCAFs in the pancreatic ductal adenocarcinoma (PDAC) TME and
   confirmed that apCAFs could induce regulatory T cell (Treg) expansion
   in ex vivo assays, although the exact mechanism remained
   unclear.[67]^17 Jiang et al. further reported that apCAF accumulation
   also depended on the interleukin (IL)-1-IL-1R2 signaling pathway in
   Tregs.[68]^18 However, their specific function in NSCLC, particularly
   in the context of NCIT, was largely unknown. Through our previously
   developed cell-communication analysis pipeline[69]^19 and a 3D ex vivo
   co-culture platform,[70]^20 we discovered that the expansion of apCAFs,
   which relied upon anti-PD1, could promote FOXP1^+ Treg accumulation via
   the PD-L2-repulsive guidance molecule b (RGMB) axis, thereby dampening
   anti-PD1 efficacy in NSCLC, in turn. Mechanistically, we discovered
   that anti-PD1 could augment apCAF expansion dependent on interferon
   (IFN)-γ, which stimulated the JAK1/2-STAT1-IFI6/27 axis, ultimately
   leading to the upregulation of MHCII and PD-L2.

   PD-L2, another PD1 ligand, showed limited efficacy with anti-PD-L2
   therapies for a long time.[71]^21 Sharpe et al. recently found RGMB as
   a new key PD-L2 partner, and combining anti-RGMB with anti-PD1 may
   overcome microbiome-mediated immunotherapy resistance.[72]^22 However,
   the specific cell cluster responsible for RGMB expression remained
   unknown. In our study, we identified Tregs as the major contributors to
   RGMB expression in the TME. Additionally, we found that apCAFs were the
   primary source of PD-L2 expression. Utilizing murine tumor models and
   organoid platforms, we confirmed that, in apCAF-enriched lung tumors,
   targeting the PD-L2-RGMB axis could significantly overcome anti-PD1
   resistance. This finding not only provides additional therapeutic
   options for addressing NCIT resistance but also opens potential
   application opportunities for anti-PD-L2 regimens in clinical settings.

Results

Antigen-presenting CAFs specifically enriched in the TME of post-non-MPR

   The TME is a complex ecosystem composed of tumor, immune, and stromal
   cells. Recently, Ye et al. identified fibroblasts as the dominant
   stromal cells subtype.[73]^23 Moreover, several studies have
   demonstrated the influence of fibroblasts on
   immunotherapy.[74]^16^,[75]^24 Coincidentally, we uncovered that higher
   expression of the CAF signature was significantly correlated with worse
   progression-free survival in a public NSCLC cohort treated with immune
   checkpoint inhibitors (ICIs) as well[76]^25 ([77]Figure S1A). To deeply
   explore fibroblasts in NCIT, we conducted scRNA-seq on tumor specimens
   from 10 NSCLC patients and DSP spatial sequencing on specimens from 6
   patients, either before or after NCIT ([78]Figure 1A). Smooth-muscle
   actin (SMA) was a typical marker of CAFs, and we therefore selected
   SMA^+ cells within the TME for DSP spatial sequencing ([79]Figure 1B).
   To comprehensively analyze the phenotypic changes within different TME
   components during NCIT treatments, we collected, sorted, and cultured
   primary CAF cell lines (37 patients), tumor-infiltrating lymphocytes
   (TILs), and organoids (11 patients). CAF cell lines were cultured in a
   3D manner in the non-immunogenic hydrogel that was developed by
   us,[80]^19^,[81]^20 followed by various readouts ([82]Figure 1A).

Figure 1.

   [83]Figure 1
   [84]Open in a new tab

   apCAFs accumulated in TME of post-non-MPR

   (A) Experimental design and the patient cohort. The respective
   experimental details referred to here are in the STAR Methods.

   (B) Three representative immunohistochemistry (IHC) pictures of
   fibrotic “hotspots” in non-small cell lung cancer (NSCLC) TME sequenced
   by DSP spatial sequencing.

   (C) Uniform manifold approximation and projection (UMAP) plot of 6,987
   CAF cells colored by clusters in our NCIT cohort. “apCAF” is encircled
   by a dashed line.

   (D) Scatterplot showing gene expression change for post-non-MPR versus
   pre-non-MPR (x axis) against the post-MPR versus pre-MPR (y axis, left)
   or post-non-MPR versus post-MPR (y axis, right) in all CAFs. Genes
   significantly upregulated in post-non-MPR versus pre-non-MPR while
   downregulated in post-MPR versus pre-MPR (left) or upregulated in
   post-non-MPR versus post-MPR (right) are marked in blue. Genes
   belonging to apCAF signature are bolded. Genes marked by red are
   well-established markers for apCAFs according to the literature.

   (E) Tumor weights of LLC tumors. n = 5 mice per group, representative
   of three independent experiments. p value is calculated using unpaired
   Student’s t test. Error bars show the mean and SEM.

   (F) Representative flow cytometric scatterplots of apCAFs within PDTF
   from non-MPR and MPR patients before or after anti-PD1 (left), with
   quantification on the right. In the barplots, apCAF means
   PDPN^+HLA-DR^+ cells (gated from CAF). Data are representative of three
   different experiments. p value is calculated using paired Student’s t
   test.

   (G) Experimental design of the PDTF platform.

   (H) Representative flow cytometric scatterplots of the transwell
   co-culture systems. Quantification barplot is on the right. Data are
   representative of three different experiments. p values are determined
   using unpaired Student’s t test, and error bars show the mean and SEM.

   During the scRNA-seq analysis, we used COL1A1 and COL3A1 as exclusive
   markers for CAFs and excluded contamination by other cell types using
   digital flow cytometry sorting (fluorescence-activated cell sorting
   [FACS]) ([85]Figures S1B–S1E). In total, we identified 6,987 CAF cells
   and categorized them into 6 distinct subtypes ([86]Figure 1C).
   Differential gene (DEG) analysis revealed unique expression patterns
   across these subtypes ([87]Figures S1F and S1G). To investigate the
   potential changes within different CAF subtypes during NCIT, we
   conducted DEG analysis systematically in all CAFs, comparing specimens
   before (pre-non-MPR and pre-MPR) and after NCIT (post-non-MPR and
   post-MPR).

   Interestingly, we found that the signature genes of apCAFs[88]^17 were
   significantly upregulated in post-non-MPR samples compared to both
   pre-non-MPR and post-MPR samples ([89]Figure 1D). Conversely, these
   genes were downregulated in post-MPR samples compared to pre-MPR
   samples ([90]Figure 1D). To validate this, we utilized another NSCLC
   NCIT dataset (GEO: [91]GSE207422
   ).[92]^26 Major subclusters identified in GEO: [93]GSE207422 were
   well-mapped to the CAF subtypes in our dataset ([94]Figures S1H and
   S1I). Due to the limited MPR patient cells in GEO: [95]GSE207422
   (73/1,063 CAFs), we focused on non-MPR patients. Besides, the fraction
   of apCAFs was larger in post-non-MPR samples compared to pre-non-MPR
   samples ([96]Figure S1J). Additionally, the expression levels of CD74,
   histocompatibility complex, class II, DR alpha (HLA-DRA) and
   histocompatibility complex, class II, DR belta 1 (HLA-DRB1) were higher
   in post-non-MPR samples compared to pre-non-MPR samples
   ([97]Figure S1K). However, whether the fraction of apCAFs was indeed
   upregulated in post-non-MPR samples remained unclear. We thus collected
   an additional NCIT cohort consisting of 14 samples for validation
   ([98]Figure S2A). By integrating the in-house supplementary dataset and
   GEO: [99]GSE207422 with our discovery dataset, we gathered an adequate
   sample size and found that apCAFs were the only subcluster to exhibit
   significant enrichment in post-non-MPR samples ([100]Figures S2A–S2C).
   Moreover, Luo et al. proposed that apCAFs likely originate from
   macrophages, and we further corroborated their findings through
   trajectory analysis, demonstrating that apCAFs indeed share
   transcriptional similarities with macrophages ([101]Figures S2D–S2G).

   During single-cell analysis, subcluster “contamination” can occur,
   although it is rare. We double-checked the expression of typical CAF
   markers on apCAFs, including Podoplanin (PDPN), COL1A1, and
   platelet-derived growth factor receptor alpha (PDGFRA)
   ([102]Figure S3A), confirming that apCAFs are indeed a distinct
   subcluster of CAFs. Additionally, we conducted bulk RNA-seq on primary
   CAF cell lines derived from patients’ tumors ([103]Figure 1A). Using
   morphological analysis and quantitative reverse-transcription PCR
   (RT-qPCR), we ensured the purity of the CAFs ([104]Figure S3B). Among
   these, there were 4 cell lines each for post-non-MPR and post-MPR. We
   created a signature by intersecting characteristic genes of apCAFs from
   the integrated single-cell dataset, naming it the “apCAF NCIT
   signature.” Indeed, the apCAF NCIT signature was significantly
   upregulated in post-non-MPR compared to post-MPR ([105]Figures S3C and
   S3D). Moreover, the apCAF NCIT signature also showed exclusive
   enrichment in post-non-MPR specimens within SMA^+ CAFs as revealed by
   DSP spatial sequencing ([106]Figures S3E and S3F).

   In summary, through multi-omics analysis, we identified a specific CAF
   subcluster, apCAFs, as a potential indicator of post-non-MPR.

apCAFs expanded during immunotherapy and dampened anti-PD1 efficacy

   Although our in silico analysis suggested that apCAFs might have
   expanded in post-non-MPR samples, its direct influence on immunotherapy
   efficacy and the factors responsible for its expansion remained
   unclear. We first conducted multiplexed immunofluorescence (mIF) to
   profile the distribution pattern of apCAFs. Notably, apCAFs were
   enriched in non-MPR samples compared to MPR samples. Moreover, after
   NCIT, the number of apCAFs was upregulated in post-non-MPR compared to
   pre-non-MPR. ([107]Figure S3G).

   Next, we constructed an orthotopic Lewis lung carcinoma (LLC) lung
   tumor model in C57BL/6 mice and sorted apCAFs and non-apCAFs from the
   tumors on day 27. These CAFs were mixed with LLC cells, followed by the
   construction of subcutaneous tumor models ([108]Figure S3H).[109]^27 We
   noticed that tumor growth curves were unaffected by non-apCAFs
   ([110]Figures 1E, [111]S3I, and S3J). Besides, apCAFs alone, without
   anti-PD1 intervention, did not exhibit pro-tumorigenic effects
   ([112]Figures 1E and [113]S3I). On the contrary, apCAFs significantly
   accelerated tumor growth during anti-PD1 regimens ([114]Figures 1E,
   [115]S3I, and S3J). Additionally, the granzyme B (GZMB) degranulation
   of CD8^+ TILs was dampened by apCAFs as well ([116]Figure S3K). This
   indicated that apCAFs could directly dampen the efficacy of anti-PD1
   therapy. Nevertheless, LLC is considered insensitive to anti-PD1
   therapy, and, in our study, anti-PD1 treatment indeed failed to induce
   significant tumor shrinkage ([117]Figures 1E and [118]S3I). Therefore,
   we utilized two additional cell lines, SJT1601 and Kras^G12DTp53^−^/^−
   (KP), both of which have been reported to respond to anti-PD1 therapy.
   Using the same methodology as in the LLC models, we demonstrated that,
   in anti-PD1-sensitive tumor models, apCAFs significantly induced
   immunotherapy resistance ([119]Figure S4). In contrast, in non-apCAF
   settings, tumor growth remained largely unaffected, regardless of
   anti-PD1 treatment ([120]Figure S4).

   However, in vivo models could not provide insights into the driving
   force behind apCAF expansion during NCIT. We then questioned whether
   apCAFs could be stimulated during anti-PD1 treatments. To explore this,
   we adopted an ex vivo 3D platform naming as patient-derived tumor
   fragments (PDTFs).[121]^28 PDTFs were encapsulated in the same hydrogel
   used in the 3D culture of CAFs, and multiplexed flow cytometry (mFC)
   was conducted on them after 48 h of anti-PD1 treatment
   ([122]Figures 1F–1H). We found that post-non-MPR samples had the
   largest fraction of apCAFs at baseline ([123]Figures 1F and [124]S5A).
   Furthermore, anti-PD1 treatment significantly enhanced apCAF expansion
   in both pre-non-MPR and post-non-MPR samples, but not in MPR samples
   ([125]Figures 1F and [126]S5A).

   To determine whether direct contact or soluble factors led to apCAF
   marker upregulation, we first sorted CAFs and non-CAFs and then used
   the transwell assay to profile changes in apCAFs ([127]Figures 1G and
   1H). Besides, non-CAFs could stimulate apCAF accumulation in a
   non-contact manner ([128]Figure 1H). Additionally, supernatants from
   non-CAFs had the same effect ([129]Figures 1F–1H). Moreover, the
   non-MPR-specific expansion of apCAFs was also effectively recapitulated
   in CAFs cultured with supernatants from non-CAFs ([130]Figure S5B).

   In summary, we concluded that anti-PD1 directly stimulated apCAF
   expansion in non-MPR samples, and apCAFs served as a direct cause of
   resistance to immunotherapy.

Anti-PD1 stimulated apCAF expansion dependent upon overexpressed IFN-γ in
non-MPR

   Although we discovered that apCAF accumulation in post-non-MPR samples
   was dependent on anti-PD1 treatment, the specific pathway responsible
   remained unclear. Re-analysis of scRNA-seq data revealed that IFN
   receptors, such as interferon alpha and beta receptor subunit 2
   (IFNAR2) (for type I IFN), interferon-gamma receptor 1/2 (IFNGR1/2)
   (for type II IFN), and IL-10RB (for type III IFN), were significantly
   upregulated in apCAFs compared to non-apCAFs ([131]Figure 2A). Pathway
   enrichment analysis confirmed that IFN-related pathways were
   upregulated in apCAFs ([132]Figure 2B), verified by gene set enrichment
   analysis (GSEA) in the NCIT dataset GEO: [133]GSE207422
   ([134]Figure 2C). Additionally, in the previously described bulk
   RNA-seq of CAFs and DSP spatial sequencing data, the IFN-γ pathway also
   showed significant enhancement in post-non-MPR samples ([135]Figures 2D
   and 2E).

Figure 2.

   [136]Figure 2
   [137]Open in a new tab

   Anti-PD1-stimulated apCAF expansion is dependent upon IFN-γ

   (A) IFN-related gene expression in apCAFs or other CAFs in our own
   NSCLC cohort.

   (B) Heatmap showing scaled gene sets scores in apCAFs and non-apCAFs in
   our own NSCLC cohort.

   (C) GSEA plot for REACTOME_INTERFERON_GAMMA_ SIGNALING in apCAFs in the
   GEO: [138]GSE207422
   cohort.

   (D and E) Scores of REACTOME_INTERFERON_GAMMA_ SIGNALING in our bulk
   RNA-seq of CAFs (D) and our DSP dataset (E). p value is calculated
   using paired Student’s t test.

   (F) The concentration of interferons (IFNs) in supernatants of PDTF.

   (G) Experimental design of the PDTF and 2D and 3D culture.

   (H) The frequency of HLA-DR^+CAFs under the conditions of PDTF and 2D
   and 3D culture (G).

   (I) Representative ridge plots of HLA-DR^+CAFs with the treatment of
   IFNs under the condition of 2D and 3D culture (upper) and
   quantification barplot (lower).

   (J) Representative immunocytochemistry (ICC) pictures of HLA-DR^+CAFs
   with different treatments. Scale bar, 50 μm.

   (K) Representative ICC pictures of HLA-DR^+CAFs with different
   treatments. Scale bar, 50 μm.

   (L) Representative flow cytometric scatterplots of CAFs with different
   treatments (left) and the quantification barplot (right).

   For (F), (H), (I), and (L), data are representative of three different
   experiments. p values are determined using unpaired Student’s t test,
   and error bars show the mean and SEM.

   To verify the in silico findings, we collected supernatants from PDTF
   culture for ELISA profiling ([139]Figures 2F and 2G). Indeed, all three
   types of IFN, including IFN-α/β, IFN-γ, and IFN-λ, showed upregulation
   in the post-non-MPR PDTF supernatants ([140]Figure 2F). We then
   transitioned from ex vivo to in vitro assays. We used the previously
   described human primary CAF biobank for downstream experiments
   ([141]Figure 2G). We found that traditional 2D culture settings largely
   lost HLA-DRA expression compared to PDTF at the baseline time point.
   Meanwhile, 3D culture of CAFs could recover HLA-DRA expression
   ([142]Figure 2H). Previous reports have also noted that CAFs lose MHCII
   molecule expression in 2D settings,[143]^29 emphasizing the necessity
   of our 3D culture system. Regardless of the culture method, IFN-γ, but
   not IFN-α/β or IFN-λ, was the major IFN responsible for apCAF
   expansion, as confirmed by immunocytochemistry (ICC) ([144]Figures 2I
   and 2J). As previously mentioned, we showed that anti-PD1 could enhance
   apCAF expansion. Using an IFN-γ neutralizing antibody (emapalumab), we
   further confirmed that IFN-γ was necessary for the stimulation of
   apCAFs by anti-PD1 ([145]Figures 2K and 2L). Supplementation with
   emapalumab largely blocked the upregulated apCAF fraction following
   anti-PD1 treatment, as corroborated by ICC ([146]Figures 2K and 2L). To
   further investigate whether IFN-γ contributes to apCAF expansion
   in vivo, we generated primary apCAFs with stable interferon-gamma
   receptor (IFNGR) knockdown (KD), negative control knockdown (NCKD)
   group and utilized the previously described co-transplantation models
   for validation ([147]Figure S5C). Compared to NCKD, KD apCAFs exhibited
   reduced expansion under anti-PD1 treatment ([148]Figure S5I). Moreover,
   KD largely abolished apCAF-mediated resistance to anti-PD1 in vivo
   ([149]Figures S5E–S5H), indicating that the IFN-γ-IFNGR axis plays a
   critical role in anti-PD1-induced apCAF expansion.

   Although we concluded that the anti-PD1-IFN-γ axis was responsible for
   the expansion of apCAFs and that apCAFs served as important mediators
   of anti-PD1 resistance, the widespread expression of IFN-γ and IFNGR1/2
   posed challenges for developing drugs targeting the “origin” of apCAFs
   due to the risk of off-target effects. Therefore, after addressing the
   questions of “what are apCAFs” and “where do they come from,” we set
   out to investigate “where do they go,” aiming to identify “targetable”
   axes centered around apCAFs.

Spatial co-localization of apCAFs with Tregs indicated their potential
interactions

   Using CellChat analysis, we found that CAFs exhibited the strongest
   cell-cell communication among major cell types within the TME
   ([150]Figure S5J). We hypothesize that apCAFs contribute to ICI
   resistance through interactions with other cell types. TILs, the
   primary ICI responders, showed stronger interaction with apCAFs than
   non-apCAFs ([151]Figure S5K), including CD8 TILs and Tregs. We
   previously proposed that non-CAF-secreted IFN-γ was responsible for
   apCAF expansion. Here, we identified a certain CD8 subtype, BAG3^+
   T cells, as the most probable source of IFN-γ ([152]Figures S5M–S5S).
   Additionally, compared to other known interaction partners of Tregs,
   such as macrophages and tumor cells, apCAFs exhibited even more
   interactions ([153]Figure S5L). Brekken et al. found that apCAFs from
   mice orthotopic PDAC tumors could lead to Tregs formation in in vitro
   assays.[154]^17 However, such immune-modulating effects have not been
   reported in other tumors or in in vivo scenarios. We hypothesized that
   apCAFs might also induce Tregs formation in the NSCLC TME. First, we
   tested if apCAFs were closely located near Tregs. Using mIF, we found
   that Tregs were more prevalent in post-non-MPR compared to pre- or
   post-MPR ([155]Figure S6A) and were in direct contact with
   antigen-presenting cells (APCs) (CD74^+ cells) ([156]Figure 3A), which
   are indispensable for Tregs differentiation.[157]^30 While apCAFs are
   considered non-classic APCs,[158]^17 the number of Tregs was positively
   correlated with the number of apCAFs ([159]Figures 3B and 3C), which is
   not the case for non-apCAFs ([160]Figure S6B). In post-non-MPR samples,
   the distances between Tregs and apCAFs were much shorter
   ([161]Figures 3D and [162]S6C), which is not observed in non-apCAFs
   either ([163]Figure S6D). Although the number of APCs within a 25 μm
   radius around Tregs showed no difference between post-non-MPR and
   post-MPR ([164]Figure 3E), the number of apCAFs and the ratios of
   apCAFs to APCs were significantly higher ([165]Figures 3F and 3G).
   Thus, among all APCs, apCAFs likely played a more important role in
   Treg formation in post-non-MPR compared to post-MPR or pre.

Figure 3.

   [166]Figure 3
   [167]Open in a new tab

   The spatial proximity of apCAFs and Tregs suggests potential
   communications between them

   (A) Representative multiplexed IF (mIF) pictures of pre, MPR, non-MPR
   lung tumors, showing the spatial distributions of APCs (CD74^+) and
   Tregs (FOXP3^+). Arrows indicate the interaction between Treg and
   apCAFs. For each group, at least 4 patients were included. Scale bar,
   50 μm.

   (B) Representative mIF pictures showing the distributions of apCAFs
   (COL1A1^+CD74^+) and Treg (FOXP3^+). For each group, at least 4
   patients were included. Arrows indicate the spatial distribution of
   Treg. Scale bar, 60 μm.

   (C) Scatterplot showing the relationships between the number of Tregs
   and apCAFs in (B), colored by groups. Only visions with comparable
   DAPI^+ cells were selected for calculation. Each dot represents a
   vision. For each group, at least 4 patients were included; for each
   patient, at least 3 visions with Tregs were selected randomly. The
   correlation coefficient is calculated by Pearson, and p value is
   determined by two-sided linear regression t test.

   (D) Violin plot showing the shortest distance of Treg to apCAFs/APC in
   (B). The shortest distance was determined by the distance between the
   apCAFs closest to a specific Treg. Only visions with comparable DAPI^+
   cells were selected for calculation. Each dot represents a Treg cell.
   For each group, at least 4 patients were included; for each patient, at
   least 3 visions with Tregs were selected randomly; for each vision,
   every Treg was designated a shortest distance value. Significance was
   determined using unpaired Student’s t test.

   (E–G) (E) Boxplots showing the numbers of APCs within 25 μm radius
   centered around Tregs in (A). Only visions with comparable DAPI^+ cells
   were selected for calculation. (F) Boxplots showing the numbers of
   apCAFs within 25 μm radius centered around Treg in (B). Data are
   representative of three different experiments. (G) Boxplots showing the
   numbers of apCAFs/APCs within 25 μm radius centered around each Treg in
   (B). For (E)–(G), each dot represents a Treg cell. For each group, at
   least 4 patients were included; for each patient, at least 3 visions
   with Tregs were selected randomly; for each vision, at least one Treg
   was adopted to calculate the value. p value is calculated using
   unpaired Student’s t test.

   (H) UMAP plot of 9,363 CD4^+T cells colored by clusters in our NSCLC
   cohort.

   (I) UMAP plot of 13,893 CD4^+T cells colored by clusters in the GEO:
   [168]GSE207422
   cohort.

   (J) Heatmap showing the correlations between CD4^+T cells in our own
   CAF cohort and GEO: [169]GSE207422
   calculated by SingleR.

   (K) The proportions of CD4^+T cell clusters. Significance was
   determined using unpaired Student’s t test, and the absence of
   significance indicates no significant difference. ∗p < 0.05,
   ∗∗p < 0.01.

   (L) The proportions of IL-1R1_Treg in different efficacy groups from
   the GEO: [170]GSE207422
   cohort. p value is calculated by unpaired Student’s t test.

   (M) Network graphs representing the relative interaction strength
   between CAFs and CD4^+ T cells (Treg and non-Treg).

   (N) Network graphs representing the interaction between CAFs (apCAFs
   and non-apCAFs) and CD4^+ T cells. Edge strength represents the
   interaction strength calculated by CellChat.

   (O) Expression of FOXP3 and IL-1R1 in Treg and Tconv in GEO:
   [171]GSE253540
   datasets (left). Expression of FOXP3 and IL-1R1 in Treg, Tconv, and
   CD8^+T in PBMC of the healthy donors or NSCLC patients from GEO:
   [172]GSE211044 datasets (middle, right). p value is calculated using
   unpaired Student’s t test.

   (P) Monocle trajectory inference for IL-1R1_Treg, LEF1_Treg, ANXA1_CD4,
   and CCR7_CD4, colored by cell states. Analysis was conducted on our own
   CAF scRNA-seq data.

   (Q) Proportion of cell state in each cluster.

   The advent of scRNA-seq has revealed significant heterogeneity in
   TME-infiltrated Tregs.[173]^31 Ley et al. identified a cytotoxic Tregs
   subset (exTreg) that promotes pro-inflammatory responses rather than
   suppressing them.[174]^32 Therefore, a detailed analysis of Tregs is
   necessary to identify the specific subtypes interacting with apCAFs. We
   closely examined the CD4^+ TILs and found that, among the seven
   well-distinguished subclusters ([175]Figures 3H, [176]S6E, and S6F),
   five were also present in the CD4^+ TILs from GEO: [177]GSE207422
   ,[178]^26 showing excellent similarity ([179]Figures 3I, 3J, and
   [180]S6G). We then found that IL-1R1_Tregs were exclusively enriched in
   non-MPR both in our dataset and in GEO: [181]GSE207422
   ([182]Figures 3K, 3L, and [183]S6H). In contrast, another Tregs
   subtype, LEF1_Tregs, showed no significant changes in either dataset
   ([184]Figures 3K and [185]S6H). Therefore, we concluded that
   IL-1R1_Tregs, not LEF1_Tregs, is an important marker of non-MPR.

   Nevertheless, the potential interactions between IL-1R1_Tregs and
   apCAFs remained unclear. Using our previously developed
   cell-communication pipeline, we found that apCAFs were the only subset
   showing enhanced communication with Tregs compared to non-Tregs
   ([186]Figure 3M). Furthermore, apCAFs demonstrated stronger
   communication with IL-1R1_Tregs compared to non-apCAFs
   ([187]Figure 3N), indicating potential interactions. IL-1R1_Tregs is a
   newly discovered suppressive CD4^+ subset.[188]^33 We uncovered, using
   publicly available datasets,[189]^34^,[190]^35 that IL-1R1 is
   specifically expressed on Tregs from NSCLC patients’ PBMC both in vitro
   and ex vivo ([191]Figure 3O). To distinguish IL-1R1_Tregs from
   LEF1_Tregs, we conducted pseudotime analysis and found that LEF1_Tregs
   is a transitional subcluster along the differentiation trajectory from
   naive CD4^+ T cells (ANXA1_CD4 and CCR7_CD4) to IL-1R1_Tregs
   ([192]Figures 3P–3Q). Interestingly, IL-1R1_Tregs was predominant in
   state 2, which showed higher expression of exhaustion markers
   ([193]Figures 3Q, [194]S6I–S6K, [195]S7A, and S7B). However, IL-1R1
   expression on Tregs was undetectable in our LLC mice models and Prlic
   et al.’s head and neck squamous cell carcinoma (HNSCC) mouse
   models,[196]^33 likely due to potential intratumoral microbiome
   colonization. This dilemma remains unresolved, prompting us to consider
   surrogate markers for IL-1R1_Tregs through other multimodal analyses to
   better investigate apCAF-Tregs relationships in vivo.

apCAF-stimulated Tregs were characterized by FOXP1 expression

   To identify markers equivalent to IL-1R1, we used single-cell
   regulatory network inference and clustering (SCENIC) analysis
   to pinpoint intrinsic transcription factors (TFs) driving IL-1R1_Tregs
   differentiation ([197]Figure S7C). Notably, as cells transitioned from
   state 3 (dominated by naive CD4^+ TILs) to state 2 (dominated by
   IL-1R1_Tregs), FOXP1 activity increased ([198]Figure 4A). Additionally,
   FOXP1 was more active in IL-1R1_Tregs compared to LEF1_Tregs
   ([199]Figure 4B) and showed significantly higher expression in
   post-non-MPR samples ([200]Figure 4C). Since FOXP1 can be reliably
   detected in mouse LLC models, we used FOXP1^+Tregs as a surrogate for
   IL-1R1_Tregs.

Figure 4.

   [201]Figure 4
   [202]Open in a new tab

   FOXP1^+Tregs specifically induced by apCAFs

   (A) SCENIC-inferred TF activity of FOXP1, FOXP3, and BATF along the
   pseudotime trajectory.

   (B) SCENIC-inferred TF activity of FOXP1 in IL-1R1_Treg and LEF1_Treg.
   p value is calculated by exact binomial test.

   (C) Scatterplot showing gene expression change for post-non-MPR versus
   pre-non-MPR (x axis) against the post-MPR versus pre-MPR (y axis, left)
   or post-non-MPR versus post-MPR (y axis, right) in Treg. Genes
   significantly upregulated in post-non-MPR versus pre-non-MPR while
   downregulated in post-MPR versus pre-MPR (left) or upregulated in
   post-non-MPR versus post-MPR (right) are marked in blue. FOXP1 is
   bolded and highlighted.

   (D) Expression of FOXP1 among immune cells from TCGA lung squamous cell
   carcinoma (LUSC) and lung adenocarcinoma (LUAD) datasets. p values are
   calculated using Wilcoxon test.

   (E) Expression of FOXP1 in different cell types in GEO: [203]GSE253540
   datasets. p value is calculated using unpaired Student’s t test.

   (F) The coherence of apCAFs and IL-1R1_Treg in ST datasets. Spots are
   colored by the average expression of their markers relatively, selected
   from single-cell dataset.

   (G) The correlation between fraction of apCAFs and IL-1R1_Treg (brown)
   or LEF1_Treg (gray) in spots from different ST datasets. The
   correlation coefficient is calculated by Pearson, and p value is
   determined by two-sided linear regression t test.

   (H) Expression heatmaps (mean values) of genes from the IL-1R1_Treg and
   stem CD4^+ T cells signature in murine intratumoral lung CD4^+ T cells,
   purified from Col1a2 Cre^ER+I-Ab^fl/fl versus I-Ab^fl/fl mice and
   analyzed by bulk RNA-seq from GEO: [204]GSE164659
   datasets.

   (I) Representative flow cytometric scatterplots of Treg in mice tumors,
   and the quantification of them, corresponding to the experiments in
   [205]Figure 1E. Data are representative of three independent
   experiments. p values are determined using unpaired Student’s t test,
   and error bars show the mean and SEM.

   (J) Representative flow cytometric scatterplots of GZMB^+CD8^+ T cells
   and the quantification of them. Data are representative of three
   different experiments. p values are determined using unpaired Student’s
   t test, and error bars show the mean and SEM.

   (K and L) Representative ridge plots of Treg induced by apCAFs or
   non-apCAFs, with the treatment of anti-PD1, and the quantification.

   For (I)–(L), data are representative of three different experiments. p
   values are determined using unpaired Student’s t test, and error bars
   show the mean and SEM.

   FOXP1 is crucial for FOXP3-mediated expression of CTLA-4 and CD25 on
   Tregs.[206]^36^,[207]^37^,[208]^38 Deleting FOXP1 in FOXP3^+ Tregs
   leads to spontaneous inflammatory diseases in mice.[209]^38 However,
   the biological roles of FOXP1^+ Tregs in tumor immunology remain
   largely unknown. First, we examined whether FOXP1 is exclusively
   expressed on CD4^+ TILs in the NSCLC TME. Deconvolution analysis of The
   Cancer Genome Atlas (TCGA) data showed that FOXP1 indeed has the
   highest expression in CD4^+ TILs ([210]Figures 4D and [211]S7D).
   Furthermore, Tregs exhibited significantly higher FOXP1 expression
   compared to conventional T cell (Tconv) or CD8^+ T cells in NSCLC PBMC
   ([212]Figure 4E). We then investigated the spatial relationship between
   FOXP1^+ Tregs and apCAFs. Using publicly available spatial
   transcriptomics datasets[213]^39^,[214]^40^,[215]^41^,[216]^42
   (ST-data), we found that the spatial distribution of apCAFs coincided
   with FOXP1^+ Tregs ([217]Figure 4F). Additionally, FOXP1^+ Treg
   fractions were more correlated with apCAF fractions in ST-data spots
   compared to LEF1^+ Tregs ([218]Figure 4G).

   We thus hypothesized that apCAFs might induce the expansion of
   FOXP1^+Tregs. Besides, compared to CD4^+ TILs from apCAF-knockout mice
   (Col1a2 CreER^+I-Ab^fl/fl), CD4^+ TILs from the control group
   (I-Ab^fl/fl) showed upregulation of FOXP1_Tregs signature genes (Foxp3,
   Il1rl1, Il1rl2, etc.) and diminished expression of stem CD4^+ T cells
   signature genes (Tcf7, Il7r, Gzmk, etc.) ([219]Figure 4H), suggesting
   that apCAFs might directly stimulate FOXP1^+ Treg expansion in vivo. In
   co-transplantation models, the TME of apCAF-enriched tumors also
   exhibited significantly more FOXP1^+ Tregs, as well as a higher overall
   number of Tregs ([220]Figures 4I and [221]S7E). Tregs sorted from the
   TME of apCAF-enriched tumors (termed “FOXP1high Tregs”) showed the
   strongest suppressive function on the cytotoxic behavior of CD8^+ TILs
   ex vivo ([222]Figures 4J and [223]S7F). We then co-cultured apCAFs or
   non-apCAFs sorted from patient-derived CAFs with corresponding
   autologous naive CD4^+ TILs. Additionally, Tregs, including
   FOXP1^+Tregs, exhibited substantial expansion in the apCAF co-culture
   groups, regardless of anti-PD1 stimulation ([224]Figures 4K and 4L).

   Taken together, we identified that FOXP1^+Tregs could well-represent
   IL-1R1^+Tregs and uncovered that apCAFs could directly stimulate
   FOXP1^+Treg expansion, both in vivo and in vitro.

Anti-PD1 induced exclusive expression of PD-L2 on apCAFs relying upon IFN-γ

   The primary objective of our investigation into the interaction between
   apCAFs and FOXP1^+ Tregs was to identify “targetable” axes to
   counteract apCAFs-mediated NCIT resistance. We first examined whether
   neutralizing HLA-DRA expression on apCAFs could inhibit Tregs
   differentiation. Besides, αMHCII only partially reduced the effect of
   apCAFs on Tregs and FOXP1^+ Tregs, indicating that additional
   mechanisms were involved ([225]Figure S8A). To explore potential
   mediators beyond MHCII, we conducted Virtual Inference of
   Protein-activity by Enriched Regulon (VIPER) analysis, a
   protein-activity quantification algorithm,[226]^43 on our single-cell
   data of CAFs. The re-clustering of CAFs based on protein activity was
   highly consistent with transcriptomic data ([227]Figures 5A and 5B).
   Crucially, VIPER-derived apCAFs uniquely expressed PD-L1 and PD-L2
   ([228]Figures 5A–5C). This finding was significant as Akbari et al.
   recently demonstrated that PD-L2 is crucial for Tregs maintenance and
   stability.[229]^44 Thus, beyond MHCII/T cell receptor signaling, the
   co-inhibitory axis of PD-L1/2 may also contribute to apCAF-mediated
   Tregs expansion. Despite the low RNA levels of PD-L1/2 and the sparse
   nature of single-cell transcriptomics, which limited the identification
   of PD-L1/L2 on apCAFs, VIPER analysis successfully highlighted their
   presence. mFC analysis confirmed that apCAFs, rather than non-apCAFs,
   were indeed the primary CAF sub-cluster expressing PD-L1/L2
   ([230]Figure 5E).

Figure 5.

   [231]Figure 5
   [232]Open in a new tab

   PD-L2 and HLA-DRA expression on apCAFs both under control of IFN-γ

   (A) UMAP plot of CAFs colored by clusters in our NSCLC cohort based on
   the VIPER algorithm.

   (B) Heatmap showing certain proteins’ activity in apCAFs and
   non-apCAFs, calculated by VIPER.

   (C) Feature plots of PDCD1LG2 and CD274 protein activity based on the
   VIPER metrics.

   (D) Expression of PDCD1LG2 among immune cells from TCGA LUSC and LUAD
   datasets. p values are calculated using Wilcoxon test.

   (E–G) (E) Representative flow cytometric histograms of expression of
   PD-L1 and PD-L2 in apCAFs and non-apCAFs, with quantification on the
   right. (F) PD1, PD-L1, and PD-L2 expression in tumor cells (A549),
   immune cells (PBMCs), and CAFs assessed by RT-qPCR. (G) mRNA expression
   level of genes quantified by RT-qPCR for organoids, TILs, or CAFs with
   treatment of anti-PD1 or its isotype antibody. For (E)–(G), data are
   representative of three independent experiments. p values are
   determined using unpaired Student’s t test, and error bars show the
   mean and SEM.

   (H) Representative flow cytometric scatterplots of PDTF from non-MPR
   and MPR patients before or after anti-PD1 (left), with quantification
   in (I). Data are representative of three independent experiments.

   (I) Frequency of PD-L2^+ apCAFs in PDTF. Data are representative of
   three different experiments. p values are determined using unpaired
   Student’s t test, and error bars show the mean and SEM.

   (J) Representative immunocytochemistry (ICC) pictures of
   HLA-DR^+CAFs/PD-L2^+CAFs with the supplement of IFN-γ, emapalumab,
   fludarabine, or ruxolitinib. Data are representative of three
   independent experiments. Scale bar, 50 μm.

   (K) Heatmap showing the mRNA expression level assessed by RT-qPCR. Each
   specific column means a specific patient’s CAF cell line. For every
   gene listed in the apCAFs and PD-L1/2 signatures, the difference
   between any two treatment groups is significant, determined by two-way
   ANOVA test.

   (L) The protein expression levels of PD-L2 and HLA-DR in
   patients-derived CAFs measured by WB (left) and the integrated density
   of them measured by ImageJ. Data of three representative patients are
   posted here. p values are determined using unpaired Student’s t test.

   We next questioned the biological value of PD-L1/2^+ CAFs. First, in
   TCGA lung cancer data, CAFs contribute comparable PD-L2 expression in
   the TME as macrophages, while this is not the case for PD-L1
   ([233]Figures 5D and [234]S8B–S8D). Additionally, we found that CAFs
   manifested the highest PD-L1/2 expression compared to tumor cells and
   immune cells ([235]Figure 5F). We then conducted RT-qPCR upon
   FACS-sorted organoids, CAFs, and TILs from the same individual. While
   anti-PD1 diminished PD-L1 expression on TILs, it significantly
   upregulated HLA-DRA and PD-L2 expression on both organoids and CAFs,
   with the latter showing a much greater increase ([236]Figure 5G). Taken
   together, we suggested that CAFs were pivotal contributors to PD-L1/2
   expression in the TME during anti-PD1 treatments.

   Theoretically, anti-PD1 blocks PD1/L1-mediated effects, leading us to
   hypothesize that, in the case of NCIT resistance, PD-L2 might be
   responsible. TCGA analysis confirmed that PD-L1 expression is
   relatively low in CAFs ([237]Figure S8D), and anti-PD1 treatment did
   not further up-regulate PD-L1 expression on CAFs ([238]Figure 5G).
   Therefore, we focused on PD-L2 expression fluctuations in apCAFs using
   the previously mentioned PDTF obtained during NCIT. Additionally,
   PD-L2^+ apCAFs were significantly enriched in post-non-MPR specimens
   ([239]Figures 5H and 5I). After anti-PD1 treatments ex vivo, their
   fractions increased further in both pre- and post-non-MPR
   ([240]Figures 5H and 5I). As mentioned earlier, PD-L2 is exclusively
   expressed by apCAFs, whose expansion depends on IFN-γ. Since IFN-γ was
   reported to be responsible for PD-L2 expression on tumor cells, we
   investigated whether this was also the case for CAFs.[241]^45
   Additionally, IFN-γ significantly upregulated HLA-DRA and PD-L2
   expression on CAFs ([242]Figure 5K). Neutralization of IFN-γ, knockdown
   of IFNGR, or inhibition of JAK1/2 or STAT1 largely rescued these
   effects ([243]Figures 5K and [244]S5D). Notably, neither other CAF
   subsets such as inflammatory CAFs (iCAFs) and myofibroblast CAFs
   (myCAFs) nor co-stimulatory molecules (CD80/86) were controlled by the
   IFN-γ signaling pathways in CAFs ([245]Figure 5K). Western blot (WB),
   ICC, and mFC analyses of patient-derived CAFs cell lines further
   corroborated our RT-qPCR findings ([246]Figures 5J, 5L, and [247]S8E).
   We further used a human fibroblast cell line (HFL1) and a murine
   fibroblast (MF) to double-check our results, which remained consistent
   ([248]Figures S8F and S8G). Altogether, we concluded that anti-PD1
   stimulated PD-L2^+apCAFs expansion in an IFN-γ-dependent manner,
   offering perspectives on apCAFs-Tregs interactions.

Reprogramming apCAFs and anti-PD-L2 (clone 3.2) overcame anti-PD1 resistance

   No detailed research has been conducted on the specific regulatory axis
   responsible for PD-L2^+apCAFs until now. We discovered that IFN-γ can
   stimulate PD-L2^+apCAFs via the JAK1/2-STAT1-IFI6/27 pathway
   ([249]Figure 6A). Through WB and RT-qPCR analyses of patient-derived
   CAF cell lines, we confirmed that IFN-γ directly upregulated the
   phosphorylation of the JAK1/2-STAT1 pathway ([250]Figures 6B and 6C).
   Ruxolitinib, a selective JAK1/2 inhibitor, significantly inhibited the
   total amount of JAK1/2 and STAT1 protein expression, as well as their
   phosphorylation ([251]Figures 6B and 6C). Similarly, fludarabine, a
   STAT1 activation inhibitor, showed the same effects ([252]Figures 6B
   and 6C). Additionally, emapalumab, an IFN-γ neutralization antibody,
   reduced JAK1/2-STAT1-IFI6/27 expression at the RNA level to some extent
   ([253]Figure 6C). We further verified these findings in the HFL1 cell
   line ([254]Figures S9A and S9B). We previously identified IFI6/27 as
   markers of apCAFs through multi-omics analysis, which
   were traditionally considered downstream responders to IFN-α/β-mediated
   STAT1 activation.[255]^46 However, few studies have reported their
   correlation with IFN-γ signaling. Using RT-qPCR and ICC, we confirmed
   that JAK1/2-STAT1 also controls the expression of IFI6/27
   ([256]Figures 6C and 6D). On the other hand, since the JAK3-STAT3 axis
   is also under IFNs’ control and receptors for IFNα/β were exclusively
   enriched in apCAFs, we questioned whether there could be bypass
   activation of STAT3 or IFNα/β receptors signaling pathways in
   PD-L2^+apCAFs. However, neither IFNAR1/2 inhibitors (anifrolumab and
   IFNAR-IN-1) nor the STAT3 inhibitor (NSC74859) could reverse the
   effects of IFN-γ ([257]Figures S9C and S9D). Finally, inhibition of the
   IFN-γ-JAK1/2-STAT1 pathway significantly downregulated PD-L2 and
   HLA-DRA expression on CAFs ([258]Figures 5J–5L). In general, we
   uncovered that the IFN-γ-JAK1/2-STAT1-IFI6/27 axis was responsible for
   PD-L2^+apCAF expansion.

Figure 6.

   [259]Figure 6
   [260]Open in a new tab

   Disruption of the JAK1/2-STAT1-IFI6/27-PD-L2 axis in apCAFs to
   compromise FOXP1^+Tregs

   (A) Diagram of mechanism of IFN-γ on apCAFs.

   (B) The protein expression levels of JAK-STAT pathway. Data are
   representative of three independent experiments.

   (C) Heatmap showing the mRNA expression level for genes from JAK-STAT
   pathway assessed by RT-qPCR. Each specific column means a specific
   patient’s CAF cell line. For every gene, the difference between any two
   treatment groups is significant, determined by two-way ANOVA test.

   (D) Representative ICC pictures of IFI6/27, p-STAT1, and p-JAK1
   pathway. Scale bar, 50 μm.

   (E) Tumor growth curves: tumor cells + apCAFs + anti-PD1 (light blue),
   tumor cells (light gray), tumor + apCAFs (gray), tumor + anti-PD1 (navy
   blue), tumor + apCAFs + anti-PD-L2 (pink), and tumor + apCAFs +
   anti-PD1 + anti-PD-L2 (magenta). Individual and mean tumor volume over
   time. n = 5 mice per group, representative of three independent
   experiments.

   (F) The tumor photos of mice in (E). n = 4 mice per group.

   (G) Representative flow cytometric scatterplots of FOXP1^+Tregs in mice
   tumors and the quantification. p values are determined using unpaired
   Student’s t test, and error bars show the mean and SEM.

   (H and I) Representative ridge plots.

   For (D)–(I), data are representative of three independent experiments.
   p values are determined using unpaired Student’s t test, and error bars
   show the mean and SEM.

   Aforementioned work led us to identify interventions targeting apCAFs.
   Considering methods targeting the IFN-γ-JAK1/2-STAT1-IFI6/27 axis may
   not be cost-effective or easily translatable ([261]Figures S10A–S10H),
   we next investigated whether anti-PD-L2 (clone 3.2) could overcome
   anti-PD1 resistance in an apCAF-enriched TME. Besides, anti-PD-L2
   significantly decelerated tumor growth even in the presence of apCAFs
   ([262]Figures 6E, 6F, [263]S11A–S11C, and S11F–S11H). Furthermore,
   anti-PD-L2, or the combination of anti-PD-L2 and anti-PD1, largely
   rescued the expansion of FOXP1^+Tregs and total Tregs ([264]Figures 6G,
   [265]S11D, and S11I), resulting in reactivation of CD8^+ TILs
   ([266]Figures S10H, [267]S11E, and S11J). Interestingly, in ex vivo
   co-culture assays, although αMHCII could diminish Treg expansion to
   some extent, it had limited effects on FOXP1^+ Tregs ([268]Figures 6H
   and 6I). In contrast, the addition of anti-PD-L2 further reduced Treg
   formation and significantly inhibited FOXP1^+Treg expansion, indicating
   that FOXP1^+Tregs might be specifically induced by the PD-L2 signaling
   pathway ([269]Figures 6H and 6I). Moreover, both in vivo and in vitro,
   anti-PD-L2 combined with anti-PD1 restored the anti-tumor capacity of
   CD8^+ TILs, which was originally dampened by the apCAF-FOXP1^+Tregs
   axis. Altogether, we concluded that, since a considerable amount of
   PD-L2 was attributed to apCAFs, apCAF-enriched TME resistant to
   anti-PD1 could significantly benefit from the addition of anti-PD-L2
   (clone 3.2).

The PD-L2-RGMB axis was indispensable for apCAFs-FOXP1^+Tregs interaction

   Recently, Sharpe et al. identified RGMB as another indispensable
   binding partner of PD-L2.[270]^22 They found that TILs from mice
   transplanted with gut microbiome from non-responders to anti-PD1 had
   higher expression of RGMB. The clone 3.2 of anti-PD-L2 blocks both
   PD-L2-RGMB and PD1-PD-L2 interactions, thus synergizing with
   anti-PD1.[271]^22 We also found that, in an apCAF-enriched TME,
   anti-PD-L2 (clone 3.2) could significantly reverse resistance to
   anti-PD1. Since these effects were accompanied by a reduction in
   FOXP1^+Tregs, we hypothesized that RGMB might be involved in
   apCAFs-FOXP1^+Tregs communication.

   Currently, the differences in RGMB expression among major clusters of
   TILs remain unclear. Interestingly, we found that CD4^+ TILs exhibited
   higher RGMB expression compared to B cells, CD8^+ T cells, or natural
   killer cells in the NSCLC TME ([272]Figure 7D; [273]Figure S12A).
   Additionally, Tregs showed significantly higher RGMB levels compared to
   Tconv and CD8^+ TILs ([274]Figure 7B). This indicates that Tregs are
   the major contributors to RGMB expression among TILs. Next, our
   analysis of publicly available RNA-seq data of FOXP1^KO Tregs not only
   confirmed that FOXP1 is essential for FOXP3 but also revealed that
   FOXP1 might positively regulate RGMB ([275]Figure 7C). We verified
   these findings through mFC analysis, which showed that CD4^+ TILs
   indeed exhibited significantly higher RGMB expression ([276]Figure 7E).
   Although apCAFs did not up-regulate RGMB expression on FOXP1^+Tregs
   ([277]Figure 7G), RGMB expression on FOXP1^+Tregs was notably higher
   than that on FOXP1^−Tregs ([278]Figure 7F). Furthermore, anti-PD1
   treatment increased the fraction of RGMB^+Tregs ([279]Figure 7G).
   Finally, we observed increased chromatin accessibility of RGMB, IL-1R1,
   and FOXP1 in Tregs compared to CD8^+ T cells or Tconv, highlighting
   their potential influence on Treg formation[280]^34 ([281]Figures 7H
   and [282]S12B).

Figure 7.

   [283]Figure 7
   [284]Open in a new tab

   Blockade of PD-L2-RGMB interaction to overcome anti-PD1 resistance both
   in vivo and in PDTF platforms

   (A) Diagram of interaction between apCAFs and Treg.

   (B) Expression of RGMB in GEO: [285]GSE253540
   datasets (left).

   (C) Expression of Foxp3 and RGMB in aTreg (activated Tregs) and rTreg
   (rested Tregs) either with knockout (KO) of FOXP1 or not in the merged
   GEO: [286]GSE118595
   and GEO: [287]GSE121251 datasets.

   (D) Expression of RGMB among immune cells from TCGA and LUAD datasets.
   p values are calculated using Wilcoxon test.

   (E) Expression of RGMB in CD4^+T cells and CD8^+T cells. Data are
   representative of three independent experiments. p values are
   determined using unpaired Student’s t test, and error bars show the
   mean and SEM.

   (F) Expression of RGMB in FOXP1^+ and FOXP1^− Tregs. Data are
   representative of three independent experiments. p values are
   determined using unpaired Student’s t test, and error bars show the
   mean and SEM.

   (G) Representative ridge plots of RGMB^+ cells (upper) and
   quantification (lower). Data are representative of three independent
   experiments. p values are determined using unpaired Student’s t test,
   and error bars show the mean and SEM.

   (H) Re-analyzation results for Foxp1, Il1r1, and RGMB in Treg, Tconv,
   or CD8^+ T cells in the GEO: [288]GSE211062
   assay for transposase-accessible chromatin by sequencing (ATAC-seq)
   dataset.

   (I) Tumor growth curves: tumor + anti-PD1 (gray), tumor apCAFs +
   anti-PD1 (orange), tumor cells + apCAFs + anti-PD-L2 + anti-RGMB
   (blue), and tumor cells + apCAFs + anti-PD-L2 + anti-RGMB (light blue).
   Individual and mean tumor volume over time. n = 5 mice per group,
   representative of three independent experiments.

   (J) The tumor photos of mice in (I). n = 4 mice per group.

   (K) The tumor weight of mice in (I). n = 4 mice per group.

   (L) Representative flow cytometric scatterplots of FOXP3^+Treg and
   FOXP1^+FOXP3^+Treg in mice tumors in (I) (left) and the quantification
   of them (right). Data are representative of three different
   experiments. p values are determined using unpaired Student’s t test,
   and error bars show the mean and SEM.

   (M and N) Representative flow cytometric scatterplots of GZMB^+CD8^+T
   cells (M) and FOXP1^+FOXP3^+Tregs (N) in PDTF (left) and the
   quantification of them (right).

   For (B)–(C), p value is calculated using unpaired Student’s t test. For
   (I)–(K), data are representative of three independent experiments, p
   values are determined using unpaired Student’s t test, and error bars
   show the mean and SEM. For (L)–(N), data are representative of three
   independent experiments, and p values are determined using unpaired
   Student’s t test and error bars show the mean and SEM.

   In the apCAF-enriched TME, anti-RGMB could also overcome resistance to
   anti-PD1 ([289]Figures 7I–7K). When combined with anti-PD1 or
   anti-PD-L2 (clone 3.2), anti-RGMB significantly reduced tumor growth
   and volumes ([290]Figures 7I–7K, [291]S13A–S13C, and S13F–S13H), as
   well as the fractions of FOXP1^+Tregs and total Tregs ([292]Figures 7L,
   [293]S12C, [294]S13D, and S13I). Besides, either in vivo or in the PDTF
   platforms, the addition of anti-RGMB enormously increased GZMB
   secretion by CD8^+ TILs ([295]Figures 7M, [296]S13E, and S13J).
   Although the shrinkage of FOXP1^+Tregs under anti-RGMB was not
   significant in vitro (Figure 7N), likely due to the absence of the gut
   microbiome, we could still conclude that anti-RGMB together with
   anti-PD1/L2 managed to successfully overcome apCAF-mediated resistance
   to anti-PD1.

Discussion

   NCIT has provided a valuable opportunity to investigate the mechanisms
   underlying immunotherapy resistance. Accordingly, we conducted a
   multi-omics analysis of NSCLC patient samples treated with NCIT. Our
   focus on CAFs led to the identification of apCAFs as key mediators of
   the resistance to NCIT.

   Although apCAFs had been previously identified, the mechanisms of their
   formation remained poorly understood. In PDAC, apCAFs were thought to
   differentiate from mesothelial cells.[297]^17 Later, Jiang et al.
   demonstrated that IL-1R2^+Tregs could maintain apCAFs by inhibiting
   IL-1-IL-1R1 signaling in colon cancers.[298]^18 However, our
   single-cell data of NSCLC revealed comparable IL-1R1 expression on both
   Tregs and apCAFs, suggesting that additional factors are involved.
   Using 3D culture settings,[299]^20 we identified IFN-γ as another
   crucial mediator of apCAF expansion. Furthermore, we clarify that the
   JAK1/2-STAT1-IFI6/27 axis is the primary pathway driving apCAF
   expansion. This finding significantly advances the understanding of
   apCAFs.

   IFNs have traditionally been considered crucial to anti-tumor
   immunity.[300]^47 On the contrary, we found that IFN-γ can reduce NCIT
   efficacy by specifically stimulating apCAFs. This discovery stemmed
   from observing increased IFNs in post-non-MPR samples, which might seem
   counterintuitive since IFNs are typically considered markers of “hot”
   tumors. However, recent studies demonstrate that “hot” tumors are
   actually more susceptible to acquired immunotherapy resistance.[301]^48
   For example, IFN-γ can promote Yes-associated protein (YAP) phase
   separation in cancer cells,[302]^49 leading to resistance to anti-PD1.
   What is more, Hellmann et al. found persistent IFN-γ signaling in the
   TME of stage IV NSCLC patients with acquired anti-PD1
   resistance.[303]^50 Our research further confirms that the IFN-γ-apCAF
   axis contributes to NCIT resistance in stage I/II NSCLC as well. Both
   our study and Hellmann et al.’s[304]^50 support the idea that a
   persistently inflamed, rather than immune-desert, TME might drive
   acquired immunotherapy resistance.

   We further identified a potential mechanism of IFN-γ-mediated
   immunotherapy resistance: the exclusive expression of PD-L2 on apCAFs.
   PD-L2^+ CAFs manifested cancer-type-specific, context-dependent roles.
   PD-L2 expressed by CD29^+FAP^+ CAFs mediate Treg retention in breast
   cancer,[305]^51 while PD-L2^+CAFs accelerate CD8^+ T cells’ apoptosis
   in melanoma.[306]^52 However, the relationship between apCAFs and PD-L2
   in NSCLC was still unclear. We revealed that apCAFs, not other CAF
   subtypes, are the primary source of PD-L2 in the NSCLC TME. Although
   anti-PD1 blocked the PD1/PD-L2 axis, the PD-L2/RGMB axis emerged to
   dampen anti-PD1 efficacy. Sharpe et al. reported that RGMB on T cells
   contributes to T cell anergy.[307]^22 We further confirmed that
   FOXP1^+Tregs are the main contributors to RGMB within TILs. FOXP1 is
   indispensable for FOXP3 chromatin accessibility and Treg
   homeostasis.[308]^36^,[309]^37^,[310]^38 However, the role of FOXP1 in
   Tregs within the context of NCIT was previously unknown. We discovered
   that FOXP1 acts as a TF marker for a post-non-MPR-responsive Treg
   subcluster, IL-1R1^+Tregs, and may positively regulate RGMB expression
   on Tregs. Interestingly, as previously reported, FOXP1 could facilitate
   CTLA-4 expression on Tregs,[311]^38 and RGMB binding enhanced its
   suppressive activity.[312]^53 Taken together, we believed that, as the
   “executioner” of apCAF-mediated resistance to NCIT, RGMB^+FOXP1^+Tregs
   warrant further investigation.

   Lastly, our research highlights pivotal translational aspects and
   future research directions. First, in co-culture assays, we found that
   inhibiting the JAK1/2-STAT1 pathway in CAFs not only reduced PD-L2
   expression but also restored CD80/86, thereby reversing FOXP1^+ Treg
   expansion through co-stimulatory signaling. This suggests another
   potential strategy to counter apCAF-mediated resistance, though further
   in vivo validation is needed. For instance, the JAK1/2 inhibitor
   ruxolitinib has shown synergy with nivolumab in phase 1/2 trials for
   NSCLC and Hodgkin’s lymphoma.[313]^54^,[314]^55 Thus, whether JAK1/2
   inhibitors may also be able to reprogram apCAFs in vivo warrants
   further investigation. Finally, we also observed enrichment of apCAFs
   in pre-non-MPR, suggesting additional mechanisms for apCAF residency
   before anti-PD1 therapies. Given that apCAF expansion is IFN-γ
   dependent, we hypothesized that certain subclusters enriched in
   pre-non-MPR with high expression of IFN-γ might contribute to it and
   identified BAG3^+CD8^+TILs as a key source. Therefore, further
   investigation into the interplay between apCAFs and BAG3^+CD8^+TILs
   will be of great significance to further target apCAFs.

Limitations of the study

   The central mechanism we identified—apCAFs driving immunotherapy
   resistance via FOXP1^+ Treg induction—requires further
   contextualization within broader immune cell interactions. The role of
   apCAFs as non-professional APCs in modulating other immune cells, such
   as CD8^+T cells and dendritic cells (DCs), remains unclear. Therefore,
   future studies using advanced approaches, such as lineage-specific
   genetically engineered models, are essential to elucidate the cell-cell
   communication between apCAFs and immune cells beyond Tregs. These
   insights could uncover novel therapeutic targets beyond the PD-L2/RGMB
   axis and further enhance our understanding of apCAF-mediated resistance
   to neo-ICIs.

Resource availability

Lead contact

   Further information and requests for resources and regents should be
   directed to and will be fulfilled by the lead contact, Dr. Ziming Li
   (liziming1980@shsmu.edu.cn).

Materials availability

   This study did not generate any new unique reagents, plasmids, or mouse
   lines. All unique/stable reagents used in this study are available from
   the [315]lead contact with a completed Materials Transfer Agreement.

Data and code availability

     * •
       All data generated or analyzed during this study are included in
       this manuscript (and its supplementary information files). Raw
       human sequencing data including RNA-seq, DSP spatial sequencing,
       and scRNA-seq in this paper will be shared by the [316]lead contact
       upon request. This paper also analyzes existing, publicly available
       data (deposited data of STAR Methods). Sequencing datasets
       generated in this article without human information are available
       publicly (deposited data of [317]key resources table).
     * •
       scRNA-seq datasets were deposited into the Zenodo Database:
       [318]https://zenodo.org/records/14249268.
     * •
       DSP spatial sequencing datasets were deposited into the Zenodo
       Database: [319]https://zenodo.org/records/14249266.
     * •
       Bulk RNA-seq of CAF cell lines were deposited into the Zenodo
       Database: [320]https://zenodo.org/records/14249260.
     * •
       This study did not generate any original code.
     * •
       Any additional information required to reanalyze the data reported
       in this paper is available from the [321]lead contact upon request.

Acknowledgments