All mice were housed in specific pathogen-free facility at Mayo Clinic. The Mayo Clinic Animal Care and Use Committee (IACUC) approved all uses in this study. Mice were purchased from Jackson Laboratories (JAX stock #000664 and #003831).

Peer-reviewed publications of in vivo mouse primary T cell genetic screens were manually identified and hand curated. The number of independent screens that identified each gene candidate was calculated and the full results appear in Supplemental Table S1 . In the table, “NA” indicates that the gene was not tested in that screen, “k + 1” indicates that the gene was tested but not found to be a significant candidate by original authors, and fields containing numbers represent a significant candidate ranking (range 0-1) within the individual screen. Genetic drivers of T cell malignancy identified in a separate genetic screen (5) were filtered out of the final overlap list. Column I is the count of screens that identified each gene, and column J is the count of screens that tested each gene plus the Collier tumor driver screen. Column K is the percent of screens that tested each gene that identified it as a significant candidate.

Rank aggregation (RA) is a process of formally combining multiple ranked lists, or base rankers, into one ranked list, or aggregated ranker (6). Data from independent screens included a list of all gene candidates tested by screen and ranked on the significance measure chosen by original authors (e.g., p-value or FDR). Specifically, genes for which the significance measure was less than 0.05 were assigned a rank from 1 to k, and genes tested but not identified as significant were assigned the rank k + 1. These independent base rankers were then aggregated using the R package, ‘TopKLists’ ( Figure 1A ) using the Genome Reference Consortium Mouse Build 38 (GRCm38) genome as the underlying space. The rank aggregated output from the TopKLists geometric analysis is in Supplemental Table S2 .

Overrepresentation analysis was performed on all rank aggregated gene candidates using g:Profiler2 v0.2.1 (7) and the gene ontology (GO) biological pathways (BP) (8) and Kyoto Encyclopedia of Genes and Genomes (KEGG) genes (9). The enrichment plot in Figure 1C was generated using GOSemSim v2.26.0 (10). Gene set enrichment analysis (GSEA) was performed using the top 100 rank aggregated gene candidates as the gene set and differential gene expression on tumor-infiltrating T cells versus peripheral T cells in treatment-naïve cancer patients from Zheng, et al. (11). Count data – including cluster, patient, source, and clonotype metadata – were downloaded from Zenodo (https://zenodo.org/record/5461803#.Y9fe4-xMH6Y). Original T cell cluster annotation was maintained, and analyses focused on 47 treatment-naive cancer patients from which both tumor and peripheral blood samples were taken. For each cluster, T cell data were bifurcated into two groups: tumor-infiltrating and non-infiltrating. Differential gene expression between tumor-infiltrating and non-infiltrating groups was determined using the Seurat (v4.3.0) FindMarkers function in R and ranked using the following equation: -log₁₀(adjusted P-value)*(average log₂FC). GSEA was performed against the top 100 ranked genes from our Aggregated Gene List using the GSEAPreranked tool (GSEA v4.1.0) from the Broad Institute (12).

Cell lines are tested annually for mycoplasma and were negative at time of last test. Cell lines were authenticated by STR analysis in 2018 (IDEXX BioResearch). EG7 cell lines were cultured in R10 medium (Gibco).

Full length EHHADH, SPRR1B, or truncated AAK1 (AAK1^ΔN125 ) cDNA was cloned into retroviral backbone MSCV (Takara) with a P2A-Thy1.1 reporter. An empty vector with an eGFP reporter was used as a control. Splenocytes from wild-type C57BL/6J mice were harvested, activated with Concanavalin A (2.5 µg/ml), and cultured in complete growth media supplemented with IL-2 (50 U/ml). HEK293T cells were transfected with pCL-ECO packaging vector and MSCV retroviral vector that overexpress the constructs indicated above. After two days in culture, the polyclonally activated splenocytes were transduced with the virus generated from the transfected HEK293T cells, and transduction efficiency were determined using Thy1.1 or eGFP reporter expression via flow cytometry. Tumor cells were resuspended in sterile PBS and 1x10⁶ EG7-OVA cells were injected subcutaneously into the rear flank(s) of 6 to 12-week-old mice. Tumor growth was monitored by caliper measurement and mice were euthanized before tumors reached 2 cm at largest diameter. Statistical methods were used to determine cohort size and researchers were not blinded. Seven days post tumor injection, 3.0x10⁶ transduced cells combined with 3.0x10⁶ empty vector with eGFP control were adoptively transferred intravenously.

Tumors, spleens, or draining lymph nodes were excised and homogenized in RPMI using a gentleMACS dissociator (Miltenyi Biotec). Tumors were further processed with an enzymatic tumor dissociation kit (Miltenyi Biotec) followed by a mouse TIL positive selection kit (Miltenyi Biotec). Cell suspensions were passed through a 70 µm mesh filter and red blood cells were lysed with ACK buffer. Single cell suspensions were labeled with Ghost Red viability dye (Tonbo) and the following antibodies (BioLegend): TCRβ (H57-597), thy1.1 (OX-7), CD4 (GK1.5), CD8α (53-6.7), and Cxcr3 (CXCR3-173). Labeled cells were fixed and counted on an Attune NxT flow cytometer (Thermo Fisher). Flow data was analyzed using FlowJo software ( Supplemental Figure S1 ). Experimental schematic was created with BioRender.

Splenocytes from WT C57BL/6J mice were transduced with either an empty vector with an eGFP reporter or AAK1^ΔN125 with a Thy1.1 reporter. These cells were then placed on top of the porous membranes of Transwells with a 5.0 µm pores (Corning HTS Transwell), and the bottom of the wells contained either media only, CXCL9, or CXCL10 (Peprotech). After incubating for 90 minutes at 37°C, CountBright counting beads (Invitrogen) were added to each well, and the cells were stained for flow cytometry. Migration was analyzed based on the number of cells that migrated to the bottom of the wells compared to the total number of cells.

EG7-OVA viability upon co-culture with activated and genetically modified T cells was measured using two complementary assays. In the first assay, EG7-OVA cells, stained with CellTrace Violet (Invitrogen) were co-cultured with transduced T cells in a flat-bottom 96 well plate for an effector to target cell ratio of 25:1 and incubated at 37°C and 5% CO2 for 4 hours. Following co-culture, cells were labeled for flow cytometry with Ghost Red viability dye (Tonbo) and thy1.1 (OX-7) (BioLegend). In the second assay, EG7-OVA cells were seeded at 10,000 cells per well on flat-bottom 96-well plates coated with Fibronectin and incubated overnight at 37C. T cells, three days post transduction, were added to tumor cells at the indicated effector-to-target ratios followed by the addition of YOYO-3 Iodide (Invitrogen). YOYO-3 fluorescence was quantified with IncuCyte ZOOM (Essen BioScience) every 4 hours for 72 hours.

Cytokine production by T cells pre-treated for 5 hours with Cell Activation Cocktail (BioLegend) and Protei Cells were then labeled with Ghost Red viability dye (Tonbo) and the following antibodies (BioLegend): thy1.1 (OX-7), CD4 (GK1.5), CD8a (53-6.7), IL-10 (JES5-16E3), and IL-4 (11B11) and IFNγ (Invitrogen, XMG1.2). Labeled cells were then quantified using an Attune NxT flow cytometer (Thermo Fisher), and subsequent analysis were performed using FlowJo software.

P values were calculated using the tests described in the individual figure legends using Graphpad Prism 9 (Graphpad Software) or R.

All data and code are available online (https://github.com/RogersLabGroup/Rank-Aggregation).

Unbiased and high-throughput forward genetic screens are instrumental in candidate identification, and we recently reported results from a genome-wide Sleeping Beauty (SB) screen to identify T cell genes regulating intratumoral T cell infiltration (4). To compare our results to similar screens, we identified five independent screens that interrogated T cell-intrinsic biology including activation, proliferation, and tissue infiltration ( Table 1 ) (13–16). This large collection of complementary studies provides a unique opportunity to increase the strength of each underlying study by rigorously combining the results. A simple overlap between all six screens was calculated by counting the number of screens that identified each individual gene as a significant candidate. Gene candidates that were identified as driving oncogenes in T cell malignancies were removed (5). This yielded 27 genes that were identified in two or more independent screens including Lag3, Dgkz, and Tnfrsf18 ( Supplemental Table S1 ). All 27 are expected to be loss-of-function, as only the SB screen could produce gain-of-function mutations by inducing expression of full or truncated proteins. While these genes are likely to improve infiltration of adoptively transferred cells, the most strongly selected gain-of-function candidates uniquely identified in the SB screen were not considered using this overlap approach. Thus, we adopted a more rigorous approach to formally combine the independent screen results.

Loss-of-function was tested in all screens, while gain-of-function (overexpression) mutations were tested in just one. Further, the coverage of each individual screen differed widely, ranging from 25 genes to whole genome, such that not all genes were tested in every screen ( Table 1 ). To compensate for the differences in coverage of the underlying screens, a rank aggregation approach was adopted. Rank aggregation is the process of integrating several ranked base lists into a single aggregated ranking, weighting relative selection strength within the base list ( Figure 1A ). Based on a recent publication benchmarking algorithmic approaches for combining partial lists (e.g., 25 genes tested) with full lists (e.g., whole genome), the R package “TopKLists” with geometric output was chosen (6). The full, aggregated gene candidate list contains 1,523 genes, ranked in descending order of selection strength ( Table 2 , Supplemental Table S2 ).

Cancer immunotherapy targets in all therapy categories, including ACT, and at all stages of development were exhaustively listed by the Cancer Research Institute and published in 2018 (30). To determine the number of novel candidate targets identified by genetic screens, the aggregated gene list was compared to the current immunotherapy targets list. Fifty-nine gene candidates were found to already be in therapeutic development ( Figure 1B ), while the remaining 1,464 represent novel candidate targets. Gene ontology enrichment analysis indicates that the novel candidates identified are significantly enriched for functions involved in T cell receptor (TCR) signaling, adhesion and migration, and T cell differentiation, indicating that activated T cell genes are well-represented ( Figure 1C ). Similarly, overrepresentation analysis with the Kyoto Encyclopedia of Genes and Genomes (KEGG) indicates the aggregated gene list is significantly enriched in T cell activation and adhesion processes ( Figure 1D ). These results are consistent with known biological processes important for T cell infiltration, supporting the validity of the genetic screen approaches.

The top 100 T cell gene candidates predicted to regulate T cell infiltration into inflamed tissues were used to define a custom “T cell infiltration” gene set for gene set enrichment analysis (GSEA). We hypothesized that the T cell infiltration gene set would be differentially expressed in intratumoral versus circulating T cells. T cell gene expression data from a publicly available human cancer patient dataset was used to compute T cell differential gene expression by location (11). Specifically, gene expression data were subsetted into tumor-infiltrating (T) or peripheral (P) locations, and differential expression was calculated for GSEA. Significantly differentially expressed genes by location were enriched for the T cell infiltration gene set (FDR <0.25) ( Figure 1E ). Enrichment varied by T cell subcluster, suggesting there may be opportunity to alter intratumoral accumulation of specific T cell subsets ( Supplemental Table S3 ).

AP2 associated kinase 1 (Aak1) was the strongest selected gene candidate in the SB screen, and Aak1 was subsequently observed to play a role in chemokine receptor endocytosis and T cell migration (4). Importantly, Aak1 remained the strongest selected gene candidate even after integrating our screen results with additional independent screens and was chosen for experimental validation. In addition, two other strongly selected genes were chosen, Sprr1b and Ehhadh. All three were uniquely identified as gain-of-function (i.e., overexpression) mutations by our SB screen (4), and none of them are currently being considered for therapeutic development ( Table 2 ).

The SB screen selected for overexpression of full-length Sprr1b and Ehhadh and truncated Aak1 (Aak1^ΔN125 ). Thus, polyclonally activated splenocytes from wild-type C57BL/6 mice stably transduced with either AAK1 ^ΔN125, SPRR1B, or EHHADH (P2A-Thy1.1 reporter) or with an empty vector (P2A-EGFP reporter) for a competitive infiltration experiment. After five days of expansion, genetically modified T cells were combined at a 1:1 ratio (gene of interest to empty vector control) and adoptively transferred into EG7-OVA tumor-bearing mice (tumor growth day seven). Seven days after adoptive transfer, gene-of-interest (thy1.1) and control (GFP) transduced T cells in tumors, tumor draining lymph nodes, and spleens were quantified by flow cytometry ( Figure 2A ).

AAK1 ^ΔN125 expression doubled the number of adoptively transferred T cells in tumors (both CD4⁺ and CD8⁺) while no significant differences in accumulation of adoptively transferred T cells were observed in the tumor draining lymph node or spleen ( Figures 2B, C , Supplementary Figure S2 ). This suggests that AAK1 ^ΔN125 selectively augments T cell trafficking and infiltration into inflamed tissues, consistent with its role in promoting CXCR3-mediated chemotaxis (4). Indeed, in vitro T cell migration toward CXCR3 ligands CXCL9 and CXCL10 was enhanced upon AAK1 ^ΔN125 expression ( Figure 2D ). In contrast, SPRR1B significantly increased accumulation of adoptively transferred T cells in both the tumor (2.8-fold) and spleen (6.6-fold). The splenic increase may indicate a role in improving engraftment and/or in vivo proliferation of adoptively transferred cells consistent with the role of SPRR1B in supporting proliferation (26), though in vitro proliferation of SPRR1B expressing cells does not appear altered. Finally, EHHADH expression significantly increased accumulation of CD8+ T cells in tumors, but not CD4+ T cells, suggesting that EHHADH expression impacts T cell subsets differently.

In addition to efficient delivery of T cells to the solid tumor microenvironment, maintaining or augmenting other T cell functions, such as cytotoxicity and cytokine production, could enhance ACT efficacy. Importantly, cytotoxic capability of T cells expressing any of the three candidates was retained ( Supplementary Figure 4 ). Expression of EHHADH resulted in increased production of IFNγ and IL-4 by CD4+ T cells, and expression of AAK1^ΔN125 modestly increased IL-4 production. Production of cytokines IFNγ, IL-4, and IL-10 by CD4+ T cells were unchanged by SPRR1B expression. Together, these results confirm that AAK1 ^ΔN125, SPRR1B, and EHHADH are good candidates to improve tissue infiltration of adoptively transferred cytotoxic T cells.