Criteria to make animal studies more relevant to treating human cancer

Abstract

Certain aspects of experimental tumor models in mice most accurately reflect the biology and immunology of cancer in patients. A survey of experimental cancer immunotherapy papers published in 2020 shows most do not achieve cancer shrinkage although treatment is initiated at an early time point after cancer cell injection, which does not reflect cancer immunotherapy in patients. Even then, few current experimental approaches eradicate the injected malignant cells, most only delay outgrowth. The value of targeting mutation-encoded tumor-specific antigens becomes increasingly evident while problems of finding normal gene-encoded tumor-associated antigens as safe, effective targets persist. It might be time to refocus on realistic experimental settings and truly cancer-specific targets. These antigens are associated with the least risk of side effects.

Introduction

The hope of cancer medicine is that various forms of immunotherapy will help to cure or at least achieve durable remissions in cancer patients. Numerous new immunological approaches have been developed in mouse models and been proposed to have potential clinical impact. Every experimental cancer model may have its advantage for answering a particular question. How can we select those therapeutic approaches that are most promising to have translational value, and what would the litmus test be? Currently, experimental evidence is commonly missing that would suggest potency and safety of a given treatment in the clinics. Discussed below are criteria that when fulfilled would provide such information.

Efficacy of experimental immunotherapies

The goal of clinical immunotherapy is to eradicate cancer in patients or significantly reduce tumor load. This goal should be reflected by the results of experimental studies claiming translational significance. Therefore, we used PubMed with the key words “immunotherapy” AND “cancer” for an unbiased selection of all articles that appeared in 2020 in journals listed in Table 1. The therapeutic efficacy was assessed for each of the 138 selected articles by using the “effect size” (E) which is determined by a ratio of linearized growth rates of treated against control tumors (for details see [1]). Examination of publications in 2020 revealed that, in the vast majority of studies, treatments failed to shrink the tumors and only slowed their growth, similarly to observations we made in 2010 (Figure 1). The words “immunotherapy” and “established” tumors had been used widely in titles and abstracts but complete remission or destruction of a fully developed tumor was rarely being achieved. Studies that observed tumor shrinkage (E > 1) targeted tumors <200 mm³ and in most studies treatment started within 10 days after tumor cell inoculation (Figure 2A). In fact, in 17 of the “immunotherapy” studies, treatment began when tumors had volumes below 2 mm³ and “prevention” might have been a more appropriate term for these studies. The failure to shrink (E < 1) an existent tumor seems to be common to various forms of immunotherapy indicating a systemic problem (Figure 3). Exceptional were two studies in Figure 1 using adoptive T cell therapy that resulted in shrinkage or rejection of tumors with an average volume of ~1 cm in diameter or ~500 mm³. These two studies [2*,3] are in line with previous experimental studies that documented the therapeutic potency of adoptive T cell transfer for well-established cancers [4–6] and translated to similar findings in the clinics [7].

Table 1.

Analyzed immunotherapy approaches.

Journal	Number of publications listed by type of treatment#
	All	CIa	CARb	ATTc	Vacc + CId	Vacce	ISf	Other
Nature1	64	28	9	4	7	8	5	3
JCI2	18	8	2	3		2	1	2
AACR3	14	8	1	4	1
PNAS	12	3		1	3	2	1	2
Cell4	10	5		2		2	1
Science5	9	3	1		3		1	1
Blood6	5		3	1				1
JoI	4	2			1			1
Total	138	57	16	15	15	14	11	10

^#Complete PUBMED search: ‘free full text[sb] AND cancer AND immunotherapy AND 2020[dp] AND “mice”[MeSH Terms] NOT “review”[Publication Type] NOT “case reports”[Publication Type] NOT “clinical study”[Publication Type] NOT “clinical trial”[Publication Type] NOT “comment”[Publication Type] NOT “consensus development conference”[Publication Type] NOT “dataset”[Publication Type] NOT “editorial”[Publication Type] NOT “letter”[Publication Type] NOT “systematic review”[Publication Type]’. We included only papers from the publishing groups above and that are also available in PubMed Central. We then excluded papers with survival data only, papers that looked into gene editing only and papers that did not perform therapy. The full list of papers is available upon request.

^aCheckpoint Inhibition

^bCAR-T cell transfer

^cAdoptive T cell transfer

^dCombinational treatment of vaccination with checkpoint inhibition

^eVaccination

^fImmunostimulation

¹Cell death & Disease, Nature, Nature Biomedical Engineering, Nature Biotechnology, Nature Communications, Nature Immunology, Nature Medicine, Nature Metabolism, Nature Nanotechnology, Scientific Reports

²Includes JCI Insights

³Cancer Research, Clinical Cancer Research, Cancer Discovery, Cancer Immunology Research

⁴Cell, Cancer Cell, Cell Chemical Biology, Cell Reports, Immunity

⁵Science Advances, Science Immunology, Science Translational Medicine

⁶Includes Blood Advances

Figure 1. Most published immunotherapies in 2020, compared to 2010, still treat tumors at small sizes and still achieve only delay of tumor outgrowth.

An effect size (E) < 1 indicates that the treated tumor grows at a delayed growth rate compared to the control. E = 1 means that tumor growth was arrested due to the treatment. An E > 1 indicates tumor regression. The effect size is expressed as a ratio of linearized tumor growth rates of treated against control groups. For more detail refer to Methods in [1]. Only publications that included a control group were investigated. If tumor area was stated, the volume was estimated by $\frac{\text{area}\frac{3}{2}}{2}$. (left) A detailed analysis was performed for 2020 publications that are available in PubMed Central (PMC) that focus on cancer immunotherapies performed in mice (n = 138) including xenograft models. Analyzed journals are detailed in Table 1. Evaluated types of treatments were checkpoint inhibition (CI), CAR T cell transfer (CAR), Adoptive T cell transfer (ATT), Vaccination (Vacc), Combination of vaccination with checkpoint inhibition (Vacc + CI), Immunostimulation (IS) and other (e.g. bispecific antibodies and oncolytic viruses). (right) Same data as published in [1] from the year of 2010 used as comparison. Evaluated types of treatments were checkpoint inhibition (CI), depletion of Treg cells (Treg dep), Adoptive T cell transfer (ATT), Vaccination (Vacc), Antibody treatment (Ab), Immunostimulation (IS) and other treatments (Other).

Figure 2. Majority of studies treat early tumors with few evaluations of tumor-specific therapies.

(A – B) Detailed analysis of publications in 2020 that focus on cancer immunotherapies and are available in PubMed Central (PMC, n = 138) including xenograft models. Analyzed journals and number of publications per type of treatment are detailed in Table 1. Evaluated types of treatments were checkpoint inhibition (CI), CAR T cell transfer (CAR), Adoptive T cell transfer (ATT), Vaccination (Vacc), Combination of vaccination with checkpoint inhibition (Vacc + CI), Immunostimulation (IS) and Other (e.g. bispecific antibodies and oncolytic viruses). (A) Most publications study therapy at a very early time after inoculation of tumor cells (average day after cancer cell inoculation when treatment started: 9 ± 9 standard deviation). (B) Number of publications analyzed by type of target used by a given therapy. Artificial targets include SIY, OVA and viral antigen SV40 large T. Unknown targets refer to general use of the immune system without the knowledge of the specific target of the immune response. This applies to studies when checkpoint inhibitors or Immunostimulatory agents are used. Tumor associated antigens (TAA) refer to targets that are also found on healthy tissue such as gp100, HER2, MART-1 and NY-ESO. Tumor-specific antigens (TSA) refer to targets that arise from tumor-specific mutations and are unique targets that are not found on healthy tissue.

Figure 3. Different types of immunotherapy have a similar average effect size below 1 in the publications analyzed, showing that they are rarely able to cause tumor regression.

Shown is the median effect size as a box-and-whisker plot among different types of immunotherapy with focus on cancer immunotherapies performed in mice analyzed from publications from the year of 2020 that are available in PubMed Central (PMC) (n = 138) including xenograft models. Number of analyzed publications per type of treatment and investigated journals are detailed in Table 1. An effect size (E) < 1 indicates that the treated tumor grows at a delayed growth rate compared to the control. E = 1 (indicated as red line) means that tumor growth was arrested due to the treatment. An E > 1 indicates tumor regression. The effect size is expressed as a ratio of linearized tumor growth rates of treated against control groups. For more detail refer to Methods in [1].Only publications that included a control group were investigated.

Appropriate use of the word “tumor”

In cancer medicine, the word tumor (Latin, swelling) implies a swelling caused by abnormal growth of benign or malignant tissue consisting of cancer embedded into stroma. However, a swelling (tumor) is also one of the cardinal signs of acute inflammation [8]. In experimental tumor immunology, such a swelling is often referred to as “palpable tumor”. Most of the volume results from acute inflammation that is caused by the death of usually the majority of cancer cells following subcutaneous inoculation. These dead cells “help” very effectively the remaining viable cancer cells to form tumors [9,10]. Most of the cross-section of the early “palpable tumors” consists of dead cancer cells, fibrin deposits and inflammatory neutrophil infiltrates with only a very thin rim of viable cancer cells. Even though signs of vascularization of this rim are noticeable as early as two days after tumor cell inoculation [11], the early vascularizing lesions do not meet the “standard pathologic criteria” of an established human solid tumor [12]. Pathologists readily spot the artefactual nature of these acute inflammatory lesions during the first 7 days after cancer cell inoculation [11]. Furthermore, the prominence of neutrophils in these early “tumors” is absent in immunologically “inflamed” or “hot” human tumors [13]. Human cancer is nearly never acutely inflamed and usually established for months or years and vascularized from its margins. Nevertheless, most experimental studies focus on treating these early acutely inflamed lesions found shortly after cancer cell inoculation (Figures 1A and 2B). The translational relevance of treating such artificial malignant lesions needs discussion and it seems that the word “tumor” is losing its meaning when being used in the abscissa of figures to indicate cancer cells analyzed in an in vitro assay [14]. After 2 weeks, the acute inflammation has subsided and the experimental tumor becomes histologically indistinguishable to the pathologist from a primary human cancer. Obviously, there is a dramatic change from the initial “tumor” to a truly established tumor resembling human cancer. Eradicating such truly established tumors or arresting their growth has translational relevance.

Appropriate size of the cancer cell population being targeted.

At time of diagnosis, many human cancers weigh at least 1 gram or are 1 cm³ in volume. Human and murine cancer cells are roughly of similar size, and direct volumetric measurements using four cancer cell lines [15] indicate that 1 cm³ could contain roughly 5 × 10⁸ malignant cells. However, the actual number of malignant cells is less because ~40 – 90 % of all cells in a tumor are nonmalignant stromal cells. Furthermore, vessels, fibers and ground substance add to the volume of an established tumor. Therefore, a more realistic estimate is ~1 × 10⁸ cancer cells per cm³ tumor tissue. Assuming a modest spontaneous mutation rate of 1 in 10⁶ divisions (1 × 10⁻⁶ /genetic locus/cell/generation) [16], then a 1 cm³ tumor will contain ~100 possible escape variants. The mutation rate of human cancer cells is variable but usually considered to be higher than the spontaneous mutation rate of non-malignant human cells. Ideally, the population of experimental cancer cells being targeted should reflect the heterogeneity of the autochthonous tumor [6]. Therapy-resistant variants frequently thwart attempts to cure cancer. Therefore, using a genetically heterogeneous unmanipulated cancer cell population rather than a cloned tumor line may lead to a more realistic assessment of how to prevent relapse of variants [17]**.

Tumor size versus duration of growth.

Several lines of evidence suggest duration of growth is a much more important variable and hurdle than tumor size. For example, the same sized TC-1 tumors can be achieved either at day 7 or at day 29 if for the latter 100 times less cancer cells are being injected [18]. Comparing both situations, the Listeria HPV16-E7 vaccine [19] readily caused the rejection of small E6/E7 expressing tumors when started at day 7 after tumor cell inoculation but failed to cause rejection when used for the same sized tumors that were established for 29 days [20]. Being able to eradicate very small but long-established cancerous lesions by immunotherapy would be critical because such lesions cause relapse when they remain after surgery [21–24]. Phase III randomized controlled studies currently miss to verify the effectiveness of various clinical approaches to therapeutic vaccination of cancer patients. Also, it would be important, therapeutic vaccines could be developed to eliminate premalignant cells and lesions, for example against the premalignant stages of HPV-related malignancies. Although vaccines very effectively prevent HPV infections [25], the same vaccines are ineffective once the E6/E7 oncogenes are inserted into the host cells. Vaccines hopefully can be designed that prevent the development of carcinoma in situ (CIN) from precursors initiated by E6/E7 [26]**. Transgenic mouse model may help to develop such vaccines not only to slow down HPV-related cancers [27] but also to eradicate the precursors. Precancerous as well as cancerous lesions much smaller than 1 cm³ are now being discovered by endoscopy and advanced imaging methods and powerful mouse models have become available [28, 29*] to evaluate such possibilities.

Appropriate controls when targeting tumor-associated antigens

Treatment with an antibody to a tumor-associated antigen (TAA), HER2/ERBB2, has helped patients with metastatic breast cancer [30] and current cancer immunotherapies focus on TAAs as targets for T cells (Figure 2B). TAAs are encoded by normal genes of the patients and therefore not only expressed by some types of cancers but also by at least some non-malignant somatic cells. The flagship of a useful TAA as a T cell target is CD19 that is highly and selectively expressed on normal and malignant B cells. CD19 is a therapeutically useful target in treatment of aggressive disseminated B cell-derived malignancies. However, nonmalignant B cells are therefore also killed by anti-CD19 CAR-T cell therapy. This destruction of normal cells is being tolerated because intravenous immunoglobulin can compensate for the loss of B cells. Such a selective expression on a normal cell population or tissue that is dispensable for the patient is a rare exception. Nonetheless, hope persists that a therapeutic window can be found for other TAAs where expression levels on nonmalignant cells is low enough to be safely targeted. However, experimental settings, that are used to imply clinical relevance of targeting TAAs without causing lethal or severe side effects on normal cells and tissues, remain problematic. Experiments can expose the width of the therapeutic window between efficacy and life threatening toxicity only when a completely syngeneic mouse model is being used, or a transgenic mouse model with an expression pattern that matches the physiological expression levels and distribution patterns of the TAA as endogenously expressed in humans. For example, anti-murine CD19 CAR-T cell therapy in mice induces long-term remission of B cell acute lymphoblastic leukemia while also causing B cell aplasia that can be tolerated [31]. Other properly controlled experiments reveal that the line between severe toxicity on normal tissues and therapeutic efficacy can be extremely small (e.g., for CEA [32] or for melanocyte related antigens [4,5,33]). Unfortunately, many other studies commonly lack toxicity controls when using patient-derived xenografts, xenografts of human cancer cell lines, or syngeneic mouse cancer cells manipulated to express a xenoantigen that may not mirror the physiological expression of it [2]*.

Unexpected patient lethality was observed in studies not preceded by animal experiments. For example, treatment of a patient with anti-HER-2 CAR-T cells caused serious adverse effects and death even though the CAR was based on the therapeutic humanized monoclonal antibody that had been used extensively in patients [34]. Lethality was also observed in patients treated with anti-MAGE-A3 TCR-T cells [35–37]. Neither of the TCRs used by the two groups had been vetted by a human thymus, a process that investigates most thoroughly each autologous TCR allowing only those T cell clones to survive that are trustworthy not to react with normal cells of the individual. Skepticism is warranted for the idea that even the best computer modeling programs or algorithms can predict all of the molecular mimicries that may be recognized by a given TCR [38]. At least currently, it is highly unlikely that an in silico program can replace the level of safety normally provided by the autologous thymus and yield a TCR safe and effective to be used in patients.

Appropriate controls when targeting tumor-specific antigens

Independent of the challenges targeting antigens on human tumors, there is clear evidence for what are the most powerful antigens leading to tumor rejection. A relatively unbiased assessment comes from an analysis of 478 patients treated with 19 different immunotherapy protocols. 135 partial remissions and 40 complete remissions were observed and their only significant correlation was a response to autologous cancer cells [39]. Thus, it was concluded that success of immunotherapy, even when using checkpoint inhibitors, depended on responses to mutant tumor-specific antigens [40]. This fully agrees with the extensive experimental research of major pioneers in tumor immunology. They found that antigens which provided effective and repeatable rejection of cancer cell were “unique”, meaning individually specific for each tumor, even when originating in the same organ and from a genetically identical animal [41–45]. Later, a somatic non-synonymous single nucleotide variant (nsSNV) causing a single amino acid substitution was found to be the genetic basis of a unique tumor-specific antigen [46]. This has been confirmed as the major genetic mechanism leading to the tumor-specific antigenicity of murine and human cancers [47]. Such mutations can be presented by each of potentially 12 different MHC class I and II antigen-presenting molecules of each individual patient. T cells and thus TCRs, targeting cancer-specific mutations on MHC class I and II can be isolated from multiple different tumors [48–52]. Such TCRs have been vetted by the thymus of the patient from whom the TCR was isolated, and should therefore carry a substantially lower risk of unexpected lethal cross-reactivities than TCRs from a mouse, affinity-matured TCRs from another patient or in silico-derived TCRs. However, there are rare important examples of shared tumor-specific antigens [53]. For example, in analyses of 450 patients with adenocarcinoma of the pancreas, 90% of them had either G->D, G->V or G->R substitutions in position 12 of the oncogene KRAS [54]. Thus, patients who share the same presenting MHC molecules and one of these three mutations have a shared tumor-specific antigen that can be recognized by the same TCR leading to off-the-shelf TCR banks [55]**. Some caution is needed given the individuality and diversity of MHC presenting molecules as well as genetic polymorphisms between patients. Which TCR clones are selected or forbidden must differ substantially between individuals unless they are identical twins. How much risk is being added when using tumor-specific TCRs from another person remains presently unknown.

Conclusions

Recent studies have shown some promising advances in immunotherapy. Currently there is much focus on using checkpoint inhibitors and vaccination strategies. Further advances will require the adherence to criteria and experimental settings that will better suggest translational relevance. Recent studies in human have confirmed the long standing notion that tumor-specific mutations are ideal rejection antigens. Ideal therefore would be an autologous tumor-specific TCR-T cell therapy which would bring cancer immunotherapy into a new era. Many open questions will have to be answered including how many TCRs will be needed and when would a target be effective. For example, mutant antigens must be targeted at pathophysiological expression levels in tumors displaying unmanipulated heterogeneity [6]. Certainly, a truly personalized T cell therapy (in the strictest sense of the word) appears currently not only as the most effective and safest option for immunotherapy but also becomes affordable and realistic given the vast progress in biotechnology and should be a major focus of animal studies that aspire to be relevant to human cancers.