Immunosuppressive CD29⁺ Treg accumulation in the liver in mice upon checkpoint inhibitor therapy

Abstract

Objective:

Liver metastases are often resistant to immune checkpoint inhibitor therapy (ICI) and portend a worse prognosis compared with metastases to other locations. Regulatory T cells (Tregs) are one of several immunosuppressive cells implicated in ICI resistance of liver tumors, but the role played by Tregs residing within the liver surrounding a tumor is unknown.

Design:

Flow cytometry and single-cell RNA sequencing were used to characterize hepatic Tregs before and after ICI therapy.

Results:

We found that the murine liver houses a Treg population that, unlike those found in other organs, is both highly proliferative and apoptotic at baseline. Upon administration of αPD-1, αPD-L1, or αCTLA4, the liver Treg population doubled regardless of the presence of an intrahepatic tumor. Remarkably, this change was not due to the preferential expansion of the subpopulation of Tregs that express PD-1. Instead, a subpopulation of CD29⁺ (Itgb1, integrin β1) Tregs, that were highly proliferative at baseline, doubled its size in response to αPD-1. Partial and full depletion of Tregs identified CD29⁺ Tregs as the prominent niche-filling subpopulation in the liver, and CD29⁺ Tregs demonstrated enhanced suppression in vitro when derived from the liver but not the spleen. We identified IL2 as a critical modulator of both CD29⁺ and CD29⁻ hepatic Tregs, but expansion of the liver Treg population with αPD-1 driven by CD29⁺ Tregs was in part IL2-independent.

Conclusion:

We propose that CD29⁺ Tregs constitute a unique subpopulation of hepatic Tregs that are primed to respond to ICI agents and mediate resistance.

Introduction

Immune checkpoint inhibitors (ICI) have demonstrated efficacy in advanced hepatocellular carcinoma (HCC), but their use in metastatic liver tumors is not established [1]. Liver metastases from multiple types of primary tumors appear to respond poorly to ICI relative to other metastatic sites, and their presence is associated with worse overall survival [2-4]. These findings implicate the liver microenvironment as a barrier to effective ICI response. Indeed, the presence of intratumoral Tregs has been demonstrated to correlate inversely with survival in both primary and metastatic liver tumors [5, 6]. Tregs present in the tumor compartment have been investigated for their role in ICI-resistance, but the role of Tregs within the microenvironment of the surrounding liver parenchyma has been relatively uncharacterized.

Here, we explored the connection between hepatic Tregs and ICI resistance in murine liver tumors. We examined the properties of liver-resident Tregs in both healthy and tumor-bearing murine livers. Tregs in the liver displayed a different turnover rate compared with Tregs from other tissues. We explored the principles that govern Treg population dynamics. We hypothesize that these mechanisms link the natural behavior of hepatic Tregs for self-renewal to preferential resistance to ICI mediated by hepatic Tregs.

Materials and Methods

Tumor models

B16-F10 GFP- and luciferase-expressing melanoma cell line (B16) were recently described [7]. 4T1 breast cancer cell line (catalog no. CRL-2539, RRID: CVCL_0125) was purchased from ATCC (Manassas, VA, USA). MC38 colon cancer cell line was a gift from Dr. Jay Berzofsky (NIH). All cell lines were confirmed to be free of mycoplasma contamination and grown to confluency. Intrahepatic tumor cell injections were performed as described in detail elsewhere [8]. In brief, cell suspensions of luciferase-expressing B16, 4T1, or MC38 cell lines were prepared and resuspended in a 1:1 mixture of PBS and Matrigel (Corning, NY, USA; catalog no.354230). Intrahepatic tumor establishment was achieved by injecting 20 μL containing 200,000 tumor cells/animal into the left lateral liver lobe. Mice were randomized prior to treatment initiation. For the subcutaneous tumor model, 500,000 MC38 cells were injected into the left flank of 8-week-old B6 females. Mice were randomized prior to treatment initiation. Tumor volume was calculated every 3 days by a blinded examiner using a caliper. At day 18, tumors were excised.

Treatments

Mice were treated intraperitoneally with antibodies and pharmacologic agents at the doses described unless otherwise specified in each figure legend. Commercial sources for all reagents are listed in Supplementary data section. Sirolimus (Rapa) and cyclosporine A (CsA) were reconstituted in DMSO, diluted in PBS, and 100 μg per mouse was injected. FTY720 was resuspended in PBS, and 20 μg per mouse was injected IP. Diphtheria toxin (DT) was diluted in PBS and injected IP at the indicated dosage and frequency. For antibiotics experiments, antibiotic cocktail (ABX, 0.5 g/L vancomycin , 0.5 g/L neomycin , and 0.5 g/L primaxin) was placed in the drinking water for one week prior to antibody injection and continued until sacrifice. Neutralization, depletion, or isotype IgG controls were performed by intraperitoneal injection of 200 μg of the following antibodies for either 3 days or at the frequencies otherwise listed in figure legends: anti-CD25, anti-CD8a, anti-PD-L1, anti-CTLA4, anti-IL2, IgG2b isotype, anti-TGFβ, IgG2b isotype, IgG2a isotype. Clodronate liposomes or control liposomes were injected IV at 100 μL one day prior to, and one day following, injection of αPD-1. Recombinant IL2 and anti-IL2 were complexed as previously described [9] and injected daily for 3 days prior to αPD-1 or IgG treatment.

Statistical analysis

Sample sizes for animal studies were guided by previous studies with similar or identical tumor models. Unless otherwise noted, all samples were individual biological replicates. Statistical analysis was performed using GraphPad Prism 9.0.1 (GraphPad Software, RRID:SCR_002798). The differences between groups were tested either by an unpaired Student’s T test, one-way ANOVA, or two-way ANOVA (with a Tukey’s multiple comparison post-test) as indicated in figure legends. p <0.05 was considered statistically significant.

Additional information about mouse strains and animal use, tissue processing and flow cytometry, suppression assay, scRNAseq analysis, public data analysis and validation of CD29⁺ Treg cluster in human data can be found as supplementary data.

Results

Intrahepatic Tregs restrain efficacy of immune checkpoint inhibitor therapy against liver tumors

We first compared the effect of αPD-1 treatment with intrahepatic versus subcutaneous MC38 tumors, which have been shown to respond to αPD-1 treatment [2, 10]. αPD-1 treatment effectively impaired growth of subcutaneous MC38 tumors (Figure 1A + B). In contrast, intrahepatically growing MC38 tumors showed minimal responses to αPD-1 treatment (Figure 1C - E). Flow cytometric analysis of hepatic immune cells revealed that resistance to ICI corresponded to an increase in Treg frequency in the livers of treated mice (Figure 1F). Next, we depleted Tregs using αCD25 antibody [11] to see if Tregs inhibited the effect of αPD-1 treatment against liver tumors (Figure 1G). αCD25 treatment rescued the effect of αPD-1 treatment against intrahepatic tumors despite only causing an incomplete depletion of hepatic Tregs at the endpoint of the experiment (Figure 1H + I). αPD-1/αCD25 combination therapy increased the ratio of PD-1⁺CD8⁺ T cells to Tregs within TIL (Figure 1K), a ratio known to be prognostic in patients receiving αPD-1 [12]. Similar results were seen in a second mouse strain BALB/c and the 4T1 tumor model (Supplementary Figure 1A - B). Next, we used FoxP3-DTR mice to remove Tregs. The depletion of Tregs with DT treatment alone impaired MC38 tumor growth, but αPD-1 treatment caused a further reduction in tumor growth in DT treated mice (Figure 1L + M). In summary, using various tumor models, we showed that Tregs ameliorate the effect of αPD-1 therapy against intrahepatic tumors.

Figure 1: Tregs restrain the efficacy of immune checkpoint inhibitor against liver tumors.

B6 mice subcutaneous (sc) (A-B, IgG n=4, αPD-1 n=5) and intrahepatic (ih) (C-F) MC38 tumors were treated with αPD-1 as indicated. (F) Treg frequency of CD4+ cells in non-tumoral tissue from liver. (G-K) Intrahepatic MC38 tumor weights from B6 mice treated with αPD-1 with or without αCD25, with (I) Treg frequency of CD4⁺ T cells from liver and spleen and (K) PD-1+CD8:Treg ratio from TIL. (L-M) Intrahepatic MC38 tumor weights from FoxP3-DTR mice treated with DT and IgG control or αPD-1, respectively. (H-K) One-way Analysis of variance (ANOVA) was used. (B, E, F, M) Unpaired Student’s t-test was used. Data are shown as mean ± s.d. For all statistical tests, ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001. Data are representative of (A-K) 2 experiments or (L-M) 1 experiment.

Checkpoint blockade treatment causes Treg proliferation and activation in liver

Next, we decided to study the response of Tregs to ICI therapy in more detail. First, we injected αPD-1 into naïve (tumor-free) mice and measured the frequency of Tregs in liver and spleen. Both absolute numbers as well as the relative frequency of Tregs nearly doubled in the liver, while the increase of Tregs in the spleen was less profound (Figure 2A - C). Further analysis over time demonstrated that the number of Tregs peaked 3-5 days after αPD-1 injection (Supplementary Figure 2A).

Figure 2: Checkpoint blockade causes Treg accumulation and activation in liver.

(A-C) Flow cytometry staining of Tregs from of B6 mice 3 days after one treatment with either IgG or αPD-1, represented by (A) representative flow cytometry plot showing frequency and (B) cell number of Tregs in the liver. (C) Ratio of change in Treg frequency of CD4 in αPD-1-treated mice in liver compared with spleen. (D-I) Activation markers on hepatic Tregs following αPD-1 treatment. (K + L) Fold change of Tregs in the livers and spleens of B6 mice treated with either αPDL1 or αCTLA4. (M) Frequency and (N + O) activation markers on hepatic Tregs from the livers of B6 mice treated with either IgG or αPD-1 five days following establishment of intrahepatic MC38 tumors. (A-O) n=4 or 5 per group. Unpaired Student’s T test was used, data are shown as mean ± s.d., ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001. Data are representative of (A-C, F-H, M-O) 2 experiments or (D-E, H-L) 1 experiment.

Next, we assessed the effect of αPD-1 on the expression of Treg activation markers that signal a suppressive phenotype. αPD-1 treatment increased expression of all markers evaluated: CD39, CD73, CTLA4, CD44, GITR, and ICOS (Figure 2D - I). We also tested other ICIs. Both αPD-L1 and αCTLA4 treatment increased the frequency of Tregs in livers from naïve mice (Figure 2K + L). Similar to what we observed in B6 mice, the expansion of Treg in the liver was observed in BALB/c mice (Supplementary Figure 2B). This effect was more pronounced in liver than in spleen (Figure 2K + L and Supplementary Figure 2B) and also seen in MC38 tumor-bearing mice (Figure 2M and Supplementary Figure 2C + D). Similar to what we observed in tumor free mice, the frequency of both CTLA4⁺ and ICOS⁺ Tregs increased upon αPD-1 treatment of intrahepatic tumor-bearing mice (Figure 2N + O). In addition, αPD-1 treatment also increased liver Tregs in mice bearing subcutaneous MC38 tumors without impairment of treatment efficacy (Supplementary Figure 2D). Finally, Treg expansion in response to αPD-1 treatment was observed in intrahepatic B16 and 4T1 tumors from B6 and BALB/c mice, respectively (Supplementary Figure 2E + F).

Next, we asked whether the increase in hepatic Treg numbers was due to an influx of Tregs into the liver. Mice were treated with the S1PR agonist FTY720 to block Treg egress from lymph nodes [13], and Treg responses to αPD-1 treatment were analyzed. However, FTY720 did not affect expansion of hepatic Tregs after αPD-1 treatment (Figure 3A). Therefore, we decided to study Treg proliferation by Ki67 staining in more detail. First, we compared the number of Ki67⁺ lymphocytes subpopulations in the liver and spleen in naïve mice. In the liver, 81% of NKT cells, 76% of Tregs, and 91% of NK cells were Ki67 positive as compared to 13% of NKT cells, 36% of Tregs and 70% of NK cells in the spleen (Figure 3B). In comparison, only 8% of CD4⁺ and CD8⁺ T cells were Ki67 positive in the spleen, and the percentage of Ki67⁺ B cells was similar in spleen and liver (Figures 3B). Annexin V staining indicated that the percentage of apoptotic Tregs in the liver was more than 4 times higher in liver than in spleen (Figures 3C), and the presence of intrahepatic tumors had no effect on the ratio of Ki67⁺ Tregs (Supplementary Figure 3A). As recently described in an HCC mouse model [14], αPD-1 treatment of naïve and tumor bearing mice further increased the number of Ki67⁺ but not Annexin V⁺ hepatic Tregs (Figures 3D-G). Additional Ly6C staining, which is a marker for quiescent Tregs [15], indicated that the majority of Tregs in naïve mice were Ly6C negative (Figure 3H). Thus, we suggest that hepatic Tregs proliferate preferentially in the liver in response to αPD-1 treatment.

Figure 3: Tregs proliferate in response to IL2 in the liver.

(A) FTY720 (20μg) with or without αPD-1 (200 μg) were injected to B6 mice and lymphocytes from liver and spleen were analyzed with flow cytometry 3 days after injection. The number of hepatic Tregs from αPD-1-treated B6 mice relative to IgG are shown for each treatment group. (B-C) Percentage of (B) Ki67+ and (C) Annexin V+ hepatic or splenic lymphocytes. Percentage of (D and F) Ki67+ and (E and G) Annexin V+ Tregs from (D-E) tumor-free and (F-G) MC38-bearing livers with or without αPD-1. (H) Percentage of Ly6C+ Tregs from tumor-free mice treated with IgG or αPD-1. (I + K) Expression of (I) CD25 on tumor-free murine Tregs and (K) IL2RA (CD25) in HCC-bearing human T cell subsets from the indicated tissues. (L) The liver’s frequency of Tregs in CD4 and (M) ratio of Tregs from tumor-free B6 mice treated with daily doses of control antibody or recombinant IL2-antibody complexes for 3 days, then harvested 3 days following the final injection. (N) Number of hepatic Tregs from mice treated with αPD-1 or IgG, with or without αIL2. (A), (K), and (N) One-way ANOVA was used. (B-C) Two-way AnOVA was used. (D-I) and (M) Unpaired student’s t test was used. Data are shown as mean ± s.d., ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001. Data are representative of 1 experiment.

Tregs express the high-affinity IL2 receptor CD25 (IL2RA) to utilize IL2 and sequester it from other T cells. Formerly used as a definition marker, CD25 expression is variable on Tregs, making its description in Treg function more appropriate. We decided to study CD25 expression as a readout for local IL2 concentration in liver and spleen [16]. Tregs from murine liver expressed CD25 at half the level of splenic Tregs, a finding that was observed by others previously [17] (Figure 3I). Re-analyzing published scRNAseq data of T cells from patients with resected liver tumors, [18] we found that human Treg expression of IL2RA (CD25) was lowest in the normal liver, and higher in blood and tumor Tregs (Figure 3K). Furthermore, this relatively low IL2RA expression may have been an underestimate, since the authors employed a CD4⁺CD25⁺ sorting strategy for Treg enrichment prior to scRNAseq and likely lost significant quantities of CD4⁺FOXP3⁺CD25⁻ hepatic Tregs during this process [18]. These observations sparked our interest in exploring the role of IL2 in controlling the hepatic Treg population.

Because Treg homeostasis and function depend on IL2 [19-21], we wondered if the presumably lower IL2 concentration (based on CD25 expression) in the hepatic microenvironment was constraining Treg expansion. Exogenous IL2 was administered to mice as previously described [9]. Liver Tregs experienced a 4-fold increase in frequency, compared with only a 2-fold increase in the spleen (Figure 3L + M). IL2 neutralizing antibody reduced the population of Tregs in the liver, and combination treatment abrogated the effect of αPD-1 (Figure 3N). Sirolimus (rapamycin), the canonical inhibitor of the mTOR pathway downstream of CD25, had a similar effect as IL2 blockade, diminishing the population of hepatic Tregs and abrogating the effect of αPD-1 (Supplementary Figure 3B). We attempted to discern the source of intrahepatic IL2. CD4 T cells expressed the highest quantity of IL2 in response to ex vivo stimulation and appeared to be a potential candidate (Supplementary Figure 3C). Inversely, depletion of CD8 and NKT cells did not change the responsiveness of hepatic Tregs to αPD-1 (Supplementary Figure 3D-E). Quantities of hepatic Tregs were also higher in tumor-free FVB/N mice that correlated with a higher expression of CD25 on Tregs (Supplementary Figure 3F-G). Thus, IL2 appears to be an important factor for Treg population size in the liver both at baseline and in response to αPD-1.

Hepatic Treg response to αPD-1 is not driven by a previously characterized subset

PD-1-deficient Tregs have been shown to display enhanced suppression, proliferation, and activation markers [22], so we decided to study whether αPD-1-induced changes in Tregs were solely mediated by blocking PD-1 on Tregs. Before treatment, only 7% of hepatic Tregs were PD-1⁺ in naïve mice (Supplementary Figure 3H). αPD-1 treatment induced elevated Ki67 of both PD-1⁺ and PD-1⁻ hepatic Tregs, and the absolute number of PD-1⁻ Tregs increased (Supplementary Figure 3I-K). Because most of the liver Tregs were PD-1⁻, the quantity of PD-1⁺ Tregs did not change after αPD-1 treatment. These results argue against a direct effect of αPD-1 treatment on Tregs but favor an indirect effect, which is also in line with the observation that other immune checkpoint inhibitors such as αCTLA4 and αPD-L1 increased liver Treg frequencies (Figure 2K + L).

The identical effect of all ICI agents on hepatic Treg expansion suggested that a specific Treg subpopulation was being preferentially affected by these inflammatory treatments. Certain Treg subpopulations described by others have been shown to display particularly potent functional suppression that is crucial for maintaining peripheral tolerance in specific tissues. RORγt⁺ Tregs are induced by commensal bacteria and have importance in maintaining gut tolerance [23], while KLRG1 defines the suppressive tisST2Treg in skin and fat [24, 25]. None of these subsets appeared to increase preferentially with αPD-1 treatment (Supplementary Figures 3L-M). CXCR3 is a marker of suppressive Tregs in spleen and lymphoid organs [26], but this was also not over-represented with αPD-1 treatment (Supplementary Figure 3N). Finally, CD49b expression has been shown to be a marker of mature Tregs in liver, skin, and vascular tissues [27]. CD49b neither was highly expressed on liver Tregs nor defined an αPD-1-responsive Treg subset (Supplementary Figure 3O). Thus, we concluded that the Treg population responsive to αPD-1 in the murine liver was not defined by a previously characterized subset.

scRNAseq identifies a novel αPD-1-responsive Treg subset

To define the hepatic Treg subpopulation most responsive to αPD-1, Tregs from the livers and spleens of Foxp3^EGFP mice that had been injected with hepatic MC38 tumors were FACS-purified and subjected to scRNAseq on the 10X Genomics platform. Following data normalization, quality filter, batch correction, and supervised removal of contaminating cells, 19,095 cells were included in the final analysis. T-stochastic neighbor embedding (tSNE) analysis resolved 9 distinct clusters (Figure 4A). A Treg signature was used to confirm expression of Cd3, Cd4, Foxp3, and ikzf2 (Helios) in all clusters (Supplementary Figure 4A). Clusters appeared to separate based on organ of origin, while cells from mice with both IgG and αPD-1 were distributed within every cluster (Supplementary Figure 4B).

Figure 4: Single cell RNA profiling of hepatic Tregs from MC38 tumor-bearing mice following αPD-1 treatment.

MC38 tumors were established for 7 days in FoxP3-GFP mice, then treated with either IgG or αPD-1 for 3 days. Mice were sacrificed on day 10. Tregs were FACS-purified and pooled based on group: Liver IgG (n=3), Liver αPD-1 (n=4), Spleen IgG (n=5), Spleen αPD-1 (n=5). (A) FACS-purified hepatic and splenic Tregs depicted by tSNE plot, resolving 8 clusters of Treg subsets. (B) Top differentially expressed genes defining each cluster. (C) Relative percentages of each Treg cluster derived from IgG or αPD-1-treated liver or spleen. (D) MA plots comparing expression of select genes from C2 with eTreg or rTreg clusters. (E-F) GSEA plots showing relative enrichment of C2 cells for the indicated gene sets. (G-M) Violin plots of relative expression of select genes within the rTreg, C2, and eTreg clusters. (N-O) Trajectory analysis using RNA velocity of Tregs from the (N) liver and (O) spleen. rTreg: resting Treg, Tfr: follicular Treg, eTreg: effector Treg, fTreg: fragile Treg. (G-M) One-way ANOVA was used. ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001. (E-F) FWER p values are displayed. Data are representative of 1 experiment.

We annotated the Treg clusters by examining their top differentially expressed genes (Figure 4A-B and Table 1). C0 consisted of effector Tregs (eTreg) with high expression of suppressive genes Lag3 and Il10, as well as other markers of enhanced activation such as Tnfrsf18, Cxcr3, Klrg1, Ctla4, and Icos. C1 consisted of resting Treg (rTreg) with high expression of Satb1, Sell, and Ly6c1. C2 was an unclassified Treg subset that expressed high levels of Itgb1 (integrin β1, CD29). C3 was a population of apoptotic Tregs evidenced by high mitochondrial genes that appeared despite stringent pre-analysis QC filtering. C4 was likely follicular Tregs (Tfr) due to spleen-predominance and activation markers (e.g. Nt5e, CD73). Notably, canonical Tfr markers Cxcr5 and Bcl6 were not present [28]. C5 was fragile-like Tregs (fTreg) with numerous upregulated interferon-responsive genes (e.g. Irf7, Ifit3, Ifitm3). fTreg are known to be interferon-experienced and display reduced suppression, a role comparable to the functionally “exhausted” phenotype in effector T cells [29]. C6 was cycling Tregs with high Mki67 and Top2a expression. C7 was an unknown subset characterized by high expression of nuclear genes with even distribution amongst all treatment and organ groups.

Table 1:

Top 25 differentially expressed genes (DEGs) within each cluster from scRNAseq.

DEG Rank	Cluster
	0	1	2	3	4	5	6	7
1	CCL5	SATB1	CRIP1	GM26917	TBC1D4	ISG15	HIST1H2AP	TNFRSF4
2	NKG7	BCL2	S100A4	NEB	ID3	PLAC8	HIST1H2AE	MIF
3	AW112010	KLF2	ITGB1	MACF1	IZUMO1R	GZMB	STMN1	HSPE1
4	BCL2A1B	LY6C1	S100A10	MALAT1	TIAM1	IFIT3	HIST1H1B	NCL
5	S100A6	RPS19	S100A6	AHNAK	SH2D1A	IFIT1	HMGB2	HSPD1
6	IL10	RPS29	VIM	LARS2	CD83	BST2	MKI67	NME1
7	CAPG	SELL	TAGLN2	XIST	TCF7	ISG20	PCLAF	EIF5A
8	GIMAP7	RPS20	LSP1	RNF213	NT5E	IFI27L2A	TOP2A	IRF8
9	PGLYRP1	RPL12	PYCARD	AU020206	EPHX1	IRF7	BIRC5	NHP2
10	ARL5A	RPL35A	EMP3	4932438A13RIK	RGS10	LY6A	HIST1H2AB	NPM1
11	TNFRSF4	RPS16	ATP1B3	SLC38A2	2310001H17RIK	IFITM3	TUBA1B	C1QBP
12	S100A4	RPS24	S100A11	MYCBP2	BE692007	ZBP1	HIST1H4D	TNFRSF9
13	SRGN	RPL5	ASS1	ANKRD11	CST7	PHF11B	H2AFZ	NOP58
14	LY6A	RPS7	KLF2	SAMD9L	TESPA1	USP18	UBE2C	PPP1R14B
15	SAMSN1	RPL8	LGALS1	CLK1	TOX	TNFRSF9	HMGN2	EIF4A1
16	S100A11	MS4A4B	EMB	CCDC88C	POU2F2	RSAD2	TUBB5	ATP5G1
17	TNFRSF18	RPL21	LGALS3	CHD3	BMYC	CXCL10	DUT	RANBP1
18	ID2	RPS27	PRR13	TNFAIP3	ART2B	ALDOA	PTMA	FBL
19	LAG3	RPS3A1	SH2D1A	WNK1	SMPDL3A	IFI206	CENPF	HSP90AA1
20	KLRG1	RPL35	TKT	CCNL1	SMAP1	IIGP1	CCNB2	PA2G4
21	FGL2	RPS28	TMSB4X	KMT2A	HIVEP3	RTP4	CDCA8	RAN
22	SERPINA3G	FOXP1	S1PR4	GM42418	EVL	LGALS3BP	SELENOH	CYCS
23	CD3G	RPS4X	ITGB7	VPS54	CD81	TRIM30A	LMNB1	DDX21
24	RILPL2	RPS18	MS4A4B	ATRX	ASAP1	SLFN5	H2AFV	GNL3
25	IZUMO1R	RPS8	TSPO	LRBA	YBX3	CTLA2A	RAN	PSME2

The composition of the annotated Treg subsets derived from each organ was assessed (Figure 4C). eTregs, fTregs, cycling Tregs, and apoptotic Tregs were all found more frequently in liver, while rTreg and Tfr were higher in spleen. In response to αPD-1, the frequencies of the Treg subpopulations changed in both the liver and spleen. In the αPD-1-treated liver, there were fewer apoptotic Tregs, rTreg, and, paradoxically, eTreg. Instead, the liver contained a higher percentage of cycling Treg and fTreg, the latter indicative of inflammatory changes with αPD-1. Interestingly, C2 displayed the largest increase in frequency with αPD-1 in the liver (Figure 4C and Supplementary Figure 4C).

We suspected that C2 could represent an αPD-1-responsive Treg subset that was responsible for the significant hepatic Treg increase seen in response to αPD-1 (Figure 2A). To learn more about this subset, we compared its transcriptomic expression to that of eTregs. Genes responsible for Treg effector activation and function such as Il10, Lag3, Tnfrsf4, Tnfrsf9, Tnfrsf18, Ctla4, and Tigit were decreased in C2 compared with eTreg (Figure 4D). Despite a 2-fold increase in C2 Tregs with αPD-1 treatment, lower expression of Pdcd1 was seen in C2 compared with eTreg, in line with our observation of increased proliferation of PD-1⁻ Tregs (Supplementary Figure 3K). Conversely, markers of resting Tregs such as Sell and Ly6c1, as well as markers of circulating Tregs S1pr1 and Ccr7, were increased compared to eTreg. Tcf7 is important for activation in Tregs [30], but other T cell lineages expressing Tcf7 are characterized by their stem-like proliferative function [31]. Tcf7 expression was higher in C2 than eTreg. GSEA demonstrated that the transcriptome of C2 was enriched for “positive regulation of lymphocyte proliferation” and “integrin mediated signaling pathway” (Figure 4E + F).

We further profiled the subpopulations for genes known to be involved in Treg function. Compared with rTreg and eTreg, C2 expressed Itgb1 at the highest level (Figure 4G). Despite being the most responsive subset to αPD-1, C2 expressed intermediate levels of Pdcd1 (Figure 4H), consistent with protein-level data on the αPD-1 response of PD1⁻ hepatic Tregs (Supplementary Figures 3L + M). Markers of effector function (Lag3, Il10, and Tnfrsf9) as well as quiescence (Sell) were also expressed in an intermediate level in C2 compared with eTreg and rTreg (Figure 4I-M). Tnfrsf9 (encoding 4-1BB) in particular has been used to trace the trajectory of Tregs as they progress from resting to activated state, [28] so we hypothesized that C2 represented an intermediate state between resting and activation. Using the RNA velocity algorithm, C2 was situated as the transition point between rTreg and eTreg in liver, but downstream of effector Tregs in spleen (Figure 4N + O). Thus, we concluded that C2 represented potentially self-renewing population with intermediate activation state and high responsiveness to αPD-1 unique in the liver.

CD29⁺ Tregs are αPD-1-responsive

We searched for a marker to define and study the C2 subset of Tregs at the protein level. Itgb1 encodes CD29 (integrin β1), described previously as a marker of circulating and cytotoxic T lymphocytes [27, 32]. First, we measured the expression of CD29 on the surface of CD4, CD8, and Tregs from the liver. Whereas <10% of CD8 and CD4 cells expressed CD29, approximately 50% of Tregs were CD29⁺ (Figure 5A), a notably higher frequency than the transcriptomically-defined Tregs in C2. Furthermore, the CD29⁺ lymphocytes had the highest Ki67 expression in each group, with CD29⁺ Tregs displaying the highest proliferation (Figure 5B). CD29⁺ CD4 and CD8 T cells also had relatively higher Annexin V levels than CD29⁻ subpopulation but lower than Annexin V in CD29⁺ Tregs (Figure 5C + D). High levels of Annexin V in Tregs observed previously were present in both CD29⁺ and CD29⁻ subpopulations. Thus, CD29 was a marker of increased proliferation and apoptosis in hepatic CD4 and CD8 T cells, whereas it was a marker of only proliferation in Tregs.

Figure 5: CD29 expression on Tregs predicts response to αPD-1.

(A) Frequency of CD29⁻ positive T lymphocytes from B6 livers. (B) Expression level of Ki67 in CD29⁺ and CD29^neg hepatic T lymphocyte subsets. (C-D) Percentage of Annexin V⁺ cells within CD29⁺ and CD29^neg hepatic T lymphocyte populations. (E) Representative flow cytometry plot and (F) quantification of percentage of Tregs and (G) Ki67 levels from liver and spleen Tregs expressing CD29, Ly6C, or double-negative (DN). (H) Percentage of PD-1+ Tregs within the subsets CD29⁺, Ly6C, or DN. (I-M) Number of Treg, Ki67 percentage and levels, and ICOS expression on the above subsets of hepatic Tregs from B6 mice treated with either IgG or αPD-1. (A-D) and (H) One-way ANOVA was used. (F-G) and (I-M) Two-way ANOVA was used. Data are shown as mean ± s.d., ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001. Data are representative of 2 experiments.

To further delineate Treg subsets using flow cytometry, we developed a panel with CD29 and Ly6C. Ly6C/CD29 staining divided CD4⁺Foxp3⁺ into three subsets in naïve mice: CD29⁺Ly6C⁻, CD29⁻Ly6C⁺ corresponding to rTregs, and CD29⁻Ly6C⁻ Tregs (double negative, DN). CD29⁺ Tregs were found at increased frequency in liver relative to spleen, while DN Tregs were higher in spleen than liver (Figure 5F). Ly6C was the same in both organs despite rTregs displaying a spleen-predominance from scRNAseq (Figure 4C), suggesting that Ly6C was specific but not sensitive for rTregs and that there were Ly6C⁻ rTregs included in the DN population. Despite CD29⁺ Tregs expressing the highest levels of Ki67 in both organs, those from the liver had significantly higher Ki67 fluorescence intensity than CD29⁺ Tregs from the spleen (Figure 5G).

Next, we tested the response of each subset to a single dose of αPD-1 in naïve mice in vivo. Although CD29⁺ Tregs had the highest PD-1⁺ percentage, almost 90% of Tregs in each subset were PD-1⁻ (Figure 5H). Following αPD-1 treatment, CD29⁺ Tregs displayed a 2-fold increase in absolute cell number, while both CD29⁻Ly6C⁺ and DN Tregs showed no significant changes (Figure 5I). CD29⁺ Tregs were highly proliferative at baseline, and αPD-1 increased both percentage and MFI of Ki67 exclusively in this subgroup (Figure 5K + L). In contrast, the activation marker ICOS was elevated in CD29⁺ as well as DN Tregs treated with αPD-1 (Figure 5M). Thus, CD29⁺ Tregs were the most proliferative subpopulation in response to αPD-1, but not the only population with increased activation.

CD29⁺ Tregs can repopulate the liver niche

Tregs can sense and respond to local population deficiencies through niche-filling, an organ-specific behavior characterized by proliferation of the local Treg population remaining after a known percentage is acutely depleted [21, 33]. We suspected that CD29⁺ Tregs were a niche-filling subpopulation due to their vigorous proliferation observed in vivo. To measure hepatic Treg niche-filling, first we performed 50% Treg depletion in female mice that were heterozygous for FoxP3-DTR-EGFP as previously described [21]. A low dose of DT (50 ng) was selected and administered by a single IP injection. The low-dose DT did not affect the population as a whole at any timepoint, indicating that a subpopulation was responsible for aggressively filling the liver niche to maintain Treg levels at a quantity necessary for suppression of autoimmunity (Supplementary Figure 5A-D). Despite no changes in the Treg population as a whole, the CD29⁺ subpopulation appeared to proliferate more than DN or CD29⁻Ly6C⁺ Tregs (Supplementary Figure 5E). The Treg population peaked at 10 days, then declined by 15 days, a characteristic overfilling and correction phenomenon seen in previous Treg niche-filling experiments [21].

We also wanted to see which Treg subpopulation refilled the 100% depleted liver by using female mice that were homozygous for the FoxP3-EGFP-DTR construct. In this case, a high dose of DT (200 ng) was selected to ensure complete depletion in the liver. Depletion was highly effective in the liver with almost no Tregs remaining day 1 following DT administration (Supplementary Figure 5F). As with 50% depletion, we observed similar niche-filling behavior with 100% depletion/maximal apoptosis, proliferation, and Treg quantity occurring at days 1, 3, and 6, respectively (Supplementary Figures 5F-H). When subsets were quantified, CD29⁺ Tregs constituted the vast majority of Tregs that re-populated the liver following depletion, both during the overfilling phase at day 6 as well as the correction phase at days 10 and 16 (Supplementary Figure 5I). The highest Ki67 levels were detected in CD29⁺ Tregs at day 3 (Supplementary Figure 5K). Thus, CD29 appeared to be a marker of the most proliferative Treg subset within the Treg population in the liver, involved in both maintenance as well as repopulating the vacated liver niche.

Hepatic CD29⁺ Tregs are suppressive and respond to IL2

CD29 was present on Tregs derived from liver and spleen, however their relative proliferation rates were different (Figure 5G). To explore functional differences between liver and splenic Tregs, we purified Tregs from the respective organs of MC38 liver tumor-bearing mice that were either CD29⁺ or CD29⁻ and performed an in vitro suppression assay. Tregs from both liver and spleen were suppressive. CD29⁺ Tregs derived from liver displayed increased suppression relative to CD29⁻ hepatic Tregs, but this difference was not seen when we compared CD29⁺ vs. CD29⁻ Tregs from spleen (Figure 6A). Suppression appeared similar between treatment groups in both CD29⁺ and CD29⁻ hepatic Tregs. Thus, we believe that CD29 marks a liver-specific Treg subset that has enhanced suppressive capacity and mediates ICI treatment failure by doubling its population in response to αPD-1, while CD29 expression did not differentiate a suppressive class of splenic Tregs.

Figure 6: CD29+ Tregs are suppressive in the liver and dependent on IL2.

(A) In vitro suppression assay. FoxP3-GFP Tregs were purified using FACS from the livers or spleens of mice bearing MC38 liver tumors treated for 3 days with either IgG or αPD-1, then co-cultured with VPD450-labeled CD4⁺GFP^neg− Responder T Cells (Tres) in the presence of irradiated APCs and αCD3 for 5 days. Suppression was calculated by gating the VPD450-low Tres cells compared with unstimulated. (B) Treg quantification following rIL2c treatment from experiment from Figure 3L, stratified by CD29 expression. (C) Quantification of hepatic Tregs from mice 3 days after treatment with combination αIL2 and αPD-1 as in Figure 3N. (A) and (C) Two-way ANOVA was used. (B) One-way ANOVA was used. Data are shown as mean ± s.d., ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001. Data are representative of (A) and (C) 2 experiments or (B) 1 experiment.

Differences in Treg population control and function are governed by organ-specific factors. We tested the impact on the Treg population of various compounds which have been reported to regulate Treg frequencies in other organs. Broad-spectrum antibiotics and αTGFβ blocking antibody were chosen due to the established importance of bacterial products [33] and peripheral differentiation [34, 35], respectively. However, liver Treg frequencies remained unaffected by these treatments, and Tregs responded equivalently to αPD-1 alone (Supplementary Figures 5L-M). We next asked whether TCR signaling could also be playing a role in Treg response to αPD-1, as PD-1 can cluster at the site of TCR and provide negative regulation of signal 1 (TCR signaling) [36]. Clodronate liposomes were used to deplete myeloid APCs, and separately, cyclosporine A was used to inhibit activation signaling downstream of signal 1. However, neither of these treatments significantly altered the frequency of hepatic Treg responses to αPD-1 (Supplementary Figures 5N-O).

Because IL2/mTOR signaling appeared to control the hepatic Treg population as a whole, we wondered if the expansion observed in CD29⁺ Tregs was primarily due to IL2 upon αPD-1 treatment. We re-analyzed data from mice administered exogenous IL2 (Figure 3L-M) and observed that CD29⁺ Tregs had expanded preferentially compared with CD29⁻ Tregs (Figure 6B). Next, we re-analyzed data from combined αIL2 and αPD-1 treatments (Figure 3N). αIL2 treatment alone reduced the populations of both CD29⁺ and CD29⁻ Tregs by about 50% after 3 days of treatment (Figure 6C). Combined administration of αPD-1 and αIL2 caused a significant increase only in the CD29⁺ Treg subpopulation but not in the CD29⁻ subpopulation. While IL2/mTOR inhibition cause a decrease in entire Treg population, CD29⁺ Tregs can regain proliferation in response to αPD1 through an IL2-independent manner.

CD29^high Tregs are more proliferative in human liver cancer

CD29⁺ Tregs were recently identified in the human GI tract [37], but have not been characterized in the liver. To translate our findings to human hepatic Tregs, we re-analyzed published scRNAseq data of immune cells derived from the tumor and liver of patients who had primary liver tumors (GSE151530) [38]. There were 902 Tregs included in the analysis, which were further subdivided into 8 clusters (Figure 7A). We developed a 6-gene signature using the top DEGs from the C2 cluster of the scRNAseq data (Crip1, S100A4, Itgb1, S100A10, S100A6, Vim) and applied it to the human dataset. Human C1 demonstrated the highest expression signature compared with all other clusters but C7 (Figure 7B-C). Thus, a subset of Tregs express Itgb1 mRNA in tumor-bearing human liver.

Figure 7: ITGB1-containing Treg subset in human liver cancer.

(A) Re-clustering of Tregs from liver samples derived from patients bearing primary liver cancer (GSE151530). (B) Expression of 6-gene signature in human Treg clusters based on top 6 DEGs from mouse C2. (C) Comparisons of 6-gene signature in (B) between C1 and all other clusters. (D) Expression of CD25 on matched human Tregs from the indicated tissues. Tregs were defined as CD45⁺CD3⁺CD4⁺FOXP3⁺. (E) Gating of CD29^high and CD29^low Treg subsets on a representative tumor sample, gated off FMO. (F) Ki67+ percentage of Tregs from the indicated tissues in either CD29^high or CD29^low subsets. (C) Unpaired Student’s t-test was used. (D) Unpaired one-way ANOVA was used. (F) Paired two-way ANOVA was used. Data are shown as mean ± s.d., ns p > 0.05, * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001.

To validate our findings on the protein level, we utilized immune cells that had been isolated from patients with HCC enrolled in a tissue collection protocol (Supplementary Table 1). Using multiplexed flow cytometry, we identified CD45⁺CD3⁺CD4⁺FOXP3⁺ human Tregs. In contrast to our reanalyzed scRNAseq data (Figure 3K), Tregs from resected HCC tumors had lower CD25 expression compared with those from matched PBMC, with Tregs from tumor-adjacent liver having an intermediate CD25 expression (Figure 7D). We then categorized Tregs as CD29^high and CD29^low (Figure 7E). Consistent with our studies in mouse, CD29 expression correlated with Ki67 positivity in human hepatic Tregs from the adjacent liver and tumor, but not in matched PBMC-derived Tregs (Figure 7F). We conclude that CD29 expression correlates with proliferation in both human and mouse Tregs in the liver.

Discussion

We studied the response to immune checkpoint blockade therapy in murine models of liver metastasis and noticed that immune checkpoint inhibitor therapy failed to impair tumor growth against various tumors when they were growing in the liver but not subcutaneously. Analysis of the immune cells in the liver revealed the unexpected proliferation and accumulation of Tregs as well as upregulation of activation markers on Tregs upon treatment with different immune checkpoint inhibitors in both tumor-bearing and tumor-free mice, which prompted us to study the response of hepatic Tregs to immune checkpoint inhibitors in more detail.

Factors affecting Treg population -- proliferation and apoptosis -- were both measured at high rates in hepatic Tregs. A reduced CD25 expression on hepatic Tregs was seen and is probably due to low IL2 levels in the liver as suggested previously [39] and a possible explanation for the high apoptosis rate. Immune checkpoint inhibitor therapy and T cell activation will increase local IL2 and thereby cause proliferation of Tregs. Interestingly, proliferation of Tregs was not specific to a previously known Treg subgroup. Instead, scRNAseq identified a subpopulation of Tregs that were Itgb1⁺ with increased susceptibility to αPD-1 treatment. Notably, the eTreg population, and not C2, had the highest expression of PD-1 transcript (Pdcd1), indicating that this was not a direct effect of αPD-1 treatment on Tregs but rather an indirect effect mediated possibly by activated T cells as described above.

ICI resistance in liver metastases is a major clinical limitation for this otherwise transformative class of systemic agents [2]. Our results suggest that Tregs, actively proliferating in the liver, are one of the earliest cells to respond to αPD-1 therapy. We propose a mechanism of resistance derived from the short-term proliferation kinetics after ICI infusion demonstrated by Tregs, whose potent immunosuppressive function can thwart the pro-inflammatory effect of targeting exhaustion markers on effector T cells. Tregs doubled after treatment with all three ICI agents tested. We suggest that this behavior occurs through a common mechanism of liver-specific microenvironmental changes disproportionately affecting CD29⁺ Tregs rather than direct checkpoint axis blockade on the surface of Treg cells. As CD29 demarcates a suppressive subset of Tregs with liver specificity, the expansion of this subset in vivo may have important implications for the immune-directed therapy of liver tumors.

There were several limitations to our study. We cannot differentiate CD29 as being functionally causative of Treg proliferation or a marker of proliferating Tregs. Based on our results, combined ICI and CD29-targeted small molecules or biologics is a reasonable treatment combination. Although there are approved biologics that can target alpha integrins or integrin heterodimers on lymphocytes, no such agents that specifically target beta integrins are approved.

In summary, here we describe a novel mechanism explaining the clinical observation that hepatic metastases are less susceptive to ICI therapy. Future studies are needed to find methods to specifically target these CD29⁺ Tregs in the liver.

What is already known on this topic

• Liver tumors demonstrate an inferior response to ICI therapy.

• Regulatory T cells restrain the efficacy of ICI therapy.

What this study adds

• Regulatory T cells expressing CD29 constitute a unique subpopulation within the liver that are highly immunosuppressive and respond strongly to ICI therapy in mice.

• CD29 positive regulatory T cells exist in the tumor microenvironment of human liver cancer.

How this study might affect research, practice, or policy

• CD29 may be used as a novel functional marker of regulatory T cells and as a biomarker for the prediction of ICI response.

• CD29-targeting agents should be investigated for the purpose of overcoming Treg-mediated ICI resistance for liver tumors.

Data availability

scRNAseq data has been deposited in the GEO database (GSE221186).